Chaos and quasi-periodicity in diffeomorphisms of the solid torus
Chaos and quasi-periodicity in diffeomorphisms of the solid torus
Chaos and quasi-periodicity in diffeomorphisms of the solid torus
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Chaos</strong> <strong>and</strong> <strong>quasi</strong>-<strong>periodicity</strong> <strong>in</strong> <strong>diffeomorphisms</strong><br />
<strong>of</strong> <strong>the</strong> <strong>solid</strong> <strong>torus</strong><br />
Henk W. Broer (1) , Carles Simó (2) , <strong>and</strong> Renato Vitolo (3)<br />
5th March 2005<br />
(1 Dept. <strong>of</strong> Ma<strong>the</strong>matics, University <strong>of</strong> Gron<strong>in</strong>gen, Blauwborgje 3, 9747 AC Gron<strong>in</strong>gen, The Ne<strong>the</strong>rl<strong>and</strong>s<br />
(2 Dept. de Matemàtica Aplicada i Anàlisi, Universitat de Barcelona, Gran Via, 585, 08007 Barcelona, Spa<strong>in</strong><br />
(3 Dip. di Mat. e Informatica, Università di Camer<strong>in</strong>o, via Madonna delle Carceri, 62032 Camer<strong>in</strong>o, Italy<br />
E-mail: broer@math.rug.nl, carles@maia.ub.es, renato.vitolo@unicam.it<br />
Abstract<br />
The Hénon family <strong>of</strong> planar maps is considered driven by <strong>the</strong> Arnol ′ d family <strong>of</strong> circle<br />
maps. This leads to a five-parameter family <strong>of</strong> skew product systems on <strong>the</strong> <strong>solid</strong><br />
<strong>torus</strong>. In this paper <strong>the</strong> dynamics <strong>of</strong> this skew product family <strong>and</strong> its perturbations are<br />
studied. It is shown that, <strong>in</strong> certa<strong>in</strong> parameter doma<strong>in</strong>s, Hénon-like strange attractors<br />
occur. The existence <strong>of</strong> <strong>quasi</strong>-periodic Hénon-like attractors is partially discussed, <strong>and</strong><br />
fur<strong>the</strong>r supported by numerical evidence.<br />
Contents<br />
1 Introduction 2<br />
1.1 Sett<strong>in</strong>g <strong>of</strong> <strong>the</strong> problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2<br />
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />
1.3 Summary <strong>and</strong> outl<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />
2 Statement <strong>of</strong> <strong>the</strong> results 9<br />
2.1 Invariant circles <strong>of</strong> saddle-type <strong>and</strong> bas<strong>in</strong>s <strong>of</strong> attraction . . . . . . . . . . . . . 9<br />
2.2 Hénon-like attractors <strong>in</strong> a family <strong>of</strong> skew product maps . . . . . . . . . . . . . 10<br />
2.3 Quasi-periodic Hénon-like attractors . . . . . . . . . . . . . . . . . . . . . . . 13<br />
2.3.1 Fur<strong>the</strong>r sett<strong>in</strong>g <strong>of</strong> <strong>the</strong> problem . . . . . . . . . . . . . . . . . . . . . . . 13<br />
2.3.2 Conjectural results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />
3 Pro<strong>of</strong>s 16<br />
3.1 Bas<strong>in</strong>s <strong>of</strong> attraction <strong>and</strong> <strong>quasi</strong>-periodic <strong>in</strong>variant circles . . . . . . . . . . . . . 16<br />
3.1.1 The Tangerman-Szewc argument generalised . . . . . . . . . . . . . . . 16<br />
3.1.2 An application <strong>of</strong> kam <strong>the</strong>ory . . . . . . . . . . . . . . . . . . . . . . . 18<br />
3.2 Hénon-like attractors do exist . . . . . . . . . . . . . . . . . . . . . . . . . . . 19<br />
3.2.1 Perturbations <strong>of</strong> multimodal families . . . . . . . . . . . . . . . . . . . 20<br />
3.2.2 Multimodal families aris<strong>in</strong>g from powers <strong>of</strong> <strong>the</strong> Logistic map . . . . . . 21<br />
4 Numerical methods, results <strong>and</strong> <strong>in</strong>terpretation 28<br />
4.1 Methods <strong>and</strong> selection <strong>of</strong> parameters . . . . . . . . . . . . . . . . . . . . . . . 28<br />
4.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29<br />
4.3 Interpretations <strong>of</strong> <strong>the</strong> numerical results . . . . . . . . . . . . . . . . . . . . . . 32<br />
1
1 Introduction<br />
S<strong>in</strong>ce <strong>the</strong> 1990’s several ma<strong>the</strong>matical characterisations have been found concern<strong>in</strong>g <strong>the</strong> structure<br />
<strong>of</strong> strange attractors <strong>in</strong> families <strong>of</strong> maps. A basic example is provided by <strong>the</strong> Hénon<br />
attractor [18], occurr<strong>in</strong>g <strong>in</strong> <strong>the</strong> family <strong>of</strong> maps<br />
H a,b :<br />
2 → 2 , (x, y) ↦→ (1 − ax 2 + y, bx), (1)<br />
where a <strong>and</strong> b are real parameters. Benedicks <strong>and</strong> Carleson [2, 3] proved that <strong>the</strong>re exists<br />
a set <strong>of</strong> parameter values S, with positive Lebesgue measure, such that for all (a, b) ∈ S<br />
<strong>the</strong> Hénon map H a,b (1) has a strange attractor co<strong>in</strong>cid<strong>in</strong>g with <strong>the</strong> closure Cl (W u (p)) <strong>of</strong><br />
<strong>the</strong> unstable manifold <strong>of</strong> a saddle fixed po<strong>in</strong>t p. Here Cl (−) denotes <strong>the</strong> topological closure.<br />
Similar techniques were <strong>the</strong>n used to prove occurrence <strong>of</strong> strange attractors <strong>in</strong> parametrised<br />
families <strong>of</strong> maps, near homocl<strong>in</strong>ic tangencies <strong>in</strong> two or higher dimensions [26, 32, 36, 39],<br />
<strong>and</strong> near tangencies <strong>in</strong> <strong>the</strong> saddle-node critical case [14]. See [42] for a general set-up to<br />
prove existence <strong>of</strong> strange attractors with one positive Lyapunov exponent <strong>in</strong> families <strong>of</strong> twodimensional<br />
maps. The strange attractors considered <strong>in</strong> <strong>the</strong>se references are called Hénonlike<br />
[14, 26, 39].<br />
1.1 Sett<strong>in</strong>g <strong>of</strong> <strong>the</strong> problem<br />
In this paper we study certa<strong>in</strong> model map families, search<strong>in</strong>g <strong>the</strong>se for Hénon-like attractors<br />
as well as for so-called <strong>quasi</strong>-periodic Hénon-like attractors. A basic model for this study is<br />
<strong>the</strong> family <strong>of</strong> maps <strong>of</strong> <strong>the</strong> <strong>solid</strong> <strong>torus</strong> 2 × ¡ 1 , where ¡ 1 = /¢ is <strong>the</strong> circle, given by<br />
⎛ ⎞<br />
x<br />
⎛<br />
⎞<br />
1 − (a + ε s<strong>in</strong>(2πθ))x 2 + y<br />
⎝y⎠ ↦→ ⎝<br />
θ<br />
bx<br />
θ + α + δ s<strong>in</strong>(2πθ)<br />
⎠ , (2)<br />
where both (ε, δ) are perturbation parameters. This map is a skew product perturbation <strong>of</strong><br />
<strong>the</strong> Hénon map (1) by <strong>the</strong> Arnol ′ d family [1]<br />
A α,δ : ¡ 1 → ¡ 1 , θ ↦→ θ + α + δ s<strong>in</strong>(2πθ). (3)<br />
<strong>of</strong> maps <strong>of</strong> ¡ 1 . First let us consider <strong>the</strong> uncoupled situation where ε = 0. The dynamics <strong>of</strong><br />
<strong>the</strong> Arnol ′ d family is globally well-known <strong>and</strong> that <strong>of</strong> <strong>the</strong> Hénon family is partially known.<br />
They are organised <strong>in</strong> <strong>the</strong> respective (α, δ)- <strong>and</strong> (a, b)-parameter planes, see Figure 1. For <strong>the</strong><br />
Arnol ′ d family <strong>in</strong> <strong>the</strong> (α, δ)-plane <strong>the</strong>re is a countable union <strong>of</strong> resonance tongues with nonempty<br />
<strong>in</strong>terior, correspond<strong>in</strong>g to hyperbolic periodic dynamics. In <strong>the</strong> complement, which<br />
is <strong>of</strong> positive measure, we f<strong>in</strong>d <strong>quasi</strong>-periodic dynamics [1, 13]. See Figure 1 (A). Similarly,<br />
for <strong>the</strong> Hénon family <strong>in</strong> <strong>the</strong> (a, b)-plane <strong>the</strong>re exists a countable union <strong>of</strong> strips <strong>of</strong> non-empty<br />
<strong>in</strong>terior correspond<strong>in</strong>g to hyperbolic periodic dynamics. In <strong>the</strong> complement a set <strong>of</strong> positive<br />
measure corresponds to strange attractors [3]. Most <strong>of</strong> <strong>the</strong> strips are extremely narrow <strong>and</strong><br />
only become visible when <strong>the</strong>y <strong>in</strong>tersect ano<strong>the</strong>r strip <strong>of</strong> <strong>the</strong> same period <strong>in</strong> such a way that<br />
a “crossroad area” is created [4]. See Figure 1 (B).<br />
Remark 1. Figure 1 is mostly obta<strong>in</strong>ed by numerical computation <strong>of</strong> Lyapunov exponents<br />
[35]. Figure 1 (B) uses <strong>the</strong> orig<strong>in</strong> as <strong>in</strong>itial po<strong>in</strong>t, which can l<strong>and</strong> ei<strong>the</strong>r <strong>in</strong> a periodic s<strong>in</strong>k, or on<br />
a strange attractor or can escape ‘to <strong>in</strong>f<strong>in</strong>ity’. Notice that, due to multistability o<strong>the</strong>r <strong>in</strong>itial<br />
po<strong>in</strong>ts can tend to different attractors. Moreover, some <strong>of</strong> <strong>the</strong> <strong>periodicity</strong> strips are connected<br />
to w<strong>in</strong>dows <strong>of</strong> s<strong>in</strong>ks <strong>of</strong> <strong>the</strong> Logistic family as this occurs for b = 0. The <strong>in</strong>terpretation <strong>of</strong> <strong>the</strong><br />
results <strong>in</strong> Figure 1 (C) is given <strong>in</strong> Section 4.<br />
2
1<br />
0.8<br />
0.6<br />
(A)<br />
0.4<br />
0.2<br />
0<br />
0 0.1 0.2 0.3 0.4 0.5<br />
0.6<br />
0.5<br />
0.4<br />
(B)<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
0.2<br />
1 1.2 1.4 1.6 1.8 2<br />
0.15<br />
(C)<br />
0.1<br />
0.05<br />
0<br />
0.2 0.25 0.3 0.35 0.4<br />
Figure 1: (A) Organisation <strong>of</strong> <strong>the</strong> (α, δ)-parameter plane <strong>of</strong> <strong>the</strong> Arnol ′ d family (3) by resonance<br />
tongues, conta<strong>in</strong><strong>in</strong>g an open set with periodic dynamics (<strong>in</strong>dicated <strong>in</strong> black). The rema<strong>in</strong><strong>in</strong>g parameter<br />
values (<strong>in</strong>dicated <strong>in</strong> white) form a nowhere dense set <strong>of</strong> positive measure with <strong>quasi</strong>-periodic<br />
dynamics. (B) Organisation <strong>of</strong> <strong>the</strong> (a, b)-parameter plane <strong>of</strong> <strong>the</strong> Hénon family (1) by strips with<br />
periodic dynamics <strong>and</strong> crossroad areas (<strong>in</strong> red). A complement <strong>of</strong> positive measure conta<strong>in</strong>s strange<br />
attractors (<strong>in</strong> green). The upper right part <strong>of</strong> <strong>the</strong> diagram (<strong>in</strong> white) corresponds to escape. (C) Diagram<br />
<strong>of</strong> map (2) <strong>in</strong> <strong>the</strong> (α, ε)-plane, for a = 1.25, b = 0.3 <strong>and</strong> δ = 0.6/(2π). Visible are: doma<strong>in</strong>s<br />
which can be <strong>in</strong>terpreted has hav<strong>in</strong>g periodic attractors (code 1, yellow), <strong>quasi</strong>-periodic attractors<br />
(code 2, blue), Hénon-like attractors (code 3, red) <strong>and</strong> <strong>quasi</strong>-periodic Hénon-like attractors (code 4,<br />
light blue). For more details see <strong>the</strong> ma<strong>in</strong> text, <strong>in</strong> particular Sections 1.3, 2.3 <strong>and</strong> 4.<br />
3
For map (2) <strong>the</strong>re are at least four comb<strong>in</strong>ations <strong>of</strong> <strong>the</strong> Arnol ′ d <strong>and</strong> Hénon families for<br />
<strong>the</strong> uncoupled case ε = 0 that correspond to parameter doma<strong>in</strong>s <strong>of</strong> positive measure.<br />
1. We start consider<strong>in</strong>g <strong>the</strong> case where <strong>the</strong> Hénon family is <strong>in</strong> a periodic attractor, so<br />
where <strong>the</strong> (maximal) Lyapunov exponent Λ H < 0.<br />
(a) In <strong>the</strong> most simple case, both constituents are <strong>in</strong> a hyperbolic periodic attractor,<br />
compare with Figures 1 (A) <strong>and</strong> (B). The correspond<strong>in</strong>g (maximal) Lyapunov<br />
exponents Λ A <strong>and</strong> Λ H are both negative. In <strong>the</strong> <strong>solid</strong> <strong>torus</strong><br />
2 ¡ × 1 this also gives<br />
a hyperbolic periodic attractor, that is persistent for |ε| ≪ 1.<br />
(b) In a second case, <strong>the</strong> Arnol ′ d family is <strong>quasi</strong>-periodic, while <strong>the</strong> Hénon family is<br />
<strong>in</strong> a periodic attractor. Now Λ A = 0, while Λ H < 0. The correspond<strong>in</strong>g uncoupled<br />
dynamics <strong>in</strong> <strong>the</strong> <strong>solid</strong> <strong>torus</strong> aga<strong>in</strong> is a normally hyperbolic <strong>quasi</strong>-periodic attractor,<br />
which by centre manifold <strong>the</strong>ory [20] <strong>and</strong> by kam <strong>the</strong>ory [5, 6] has certa<strong>in</strong><br />
persistence properties for |ε| ≪ 1.<br />
2. In <strong>the</strong> two rema<strong>in</strong><strong>in</strong>g cases <strong>the</strong> Hénon family is <strong>in</strong> a strange attractor, so with Λ H > 0.<br />
This attractor is <strong>the</strong> closure Cl (W u (Orb(p))) <strong>of</strong> <strong>the</strong> unstable manifold <strong>of</strong> a periodic<br />
saddle po<strong>in</strong>t. (Below we shall be more precise.) We have to dist<strong>in</strong>guish two cases.<br />
¡ ¡<br />
(a) In <strong>the</strong> former <strong>of</strong> <strong>the</strong>se, <strong>the</strong> Arnol ′ d family is <strong>in</strong> a periodic attractor, so with Λ A < 0,<br />
<strong>and</strong> <strong>the</strong> product system has a Hénon-like attractor. It is <strong>the</strong> ma<strong>in</strong> aim <strong>of</strong> this paper<br />
to show <strong>the</strong> persistence <strong>of</strong> this attractor for |ε| ≪ 1. For illustrations see Figure<br />
2. Here we shall focus on small values <strong>of</strong> b, which allows us to rescale our model<br />
(2) by ε. In fact we shall consider a sufficiently smooth family <strong>of</strong> skew-product<br />
<strong>diffeomorphisms</strong> T α,δ,a,ε given by<br />
⎛ ⎞<br />
x<br />
⎛<br />
⎞<br />
1 − ax 2 + εf<br />
T α,δ,a,ε :<br />
2 × 1 → 2 × 1 , ⎝y⎠ ↦→ ⎝<br />
θ<br />
εg<br />
A α,δ (θ)<br />
⎠ . (4)<br />
Here (α, δ, a, ε) are parameters, while f <strong>and</strong> g are functions <strong>of</strong> (a, x, y, θ, ε, α, δ).<br />
For 0 ≤ δ < (1/2π) <strong>and</strong> α ∈ [0, 1], <strong>the</strong> map A α,δ is a diffeomorphism <strong>of</strong> <strong>the</strong> circle<br />
¡ 1 . We perturb away from cases where (α, δ) is <strong>in</strong> one <strong>of</strong> <strong>the</strong> resonance tongues,<br />
see Figure 1 (A).<br />
(b) In <strong>the</strong> latter case, where <strong>the</strong> Arnol ′ d family is <strong>quasi</strong>-periodic, so with Λ A = 0, <strong>the</strong><br />
uncoupled product dynamics is <strong>quasi</strong>-periodic Hénon-like, i.e., on an attractor <strong>of</strong><br />
<strong>the</strong> form Cl (W u (C )) , where C is a <strong>quasi</strong>-periodic <strong>in</strong>variant circle <strong>of</strong> saddle-type,<br />
aga<strong>in</strong> compare Figure 1 (A). We conjecture that this phenomenon is persistent for<br />
|ε| ≪ 1, but have only partial results <strong>in</strong> this direction, supported by numerics.<br />
For illustrations see Figures 3 <strong>and</strong> 4.<br />
The Lyapunov diagram <strong>in</strong> Figure 1 (C) strongly suggests that all four cases occur <strong>in</strong> parameter<br />
sets <strong>of</strong> positive measure. More concretely, case 1(a) corresponds to code 1; case 1(b) to code<br />
2; case 2(a) to code 3, <strong>and</strong> case 2(b) to code 4.<br />
Our <strong>in</strong>terest is with phenomena that are persistent under small perturbations, both with<strong>in</strong><br />
<strong>the</strong> skew product sett<strong>in</strong>g <strong>and</strong> beyond this. To this end, we also consider a more general class<br />
<strong>of</strong> families def<strong>in</strong>ed as follows. First let<br />
K = (K 1 , K 2 ) :<br />
2 → 2 (5)<br />
4
1<br />
θ<br />
(A)<br />
1<br />
θ<br />
(B)<br />
replacements<br />
0.5<br />
PSfrag replacements<br />
0.5<br />
(A)<br />
0<br />
-1<br />
0<br />
x<br />
1<br />
-0.4<br />
0<br />
y<br />
0.4<br />
0<br />
-1<br />
0<br />
x<br />
1 -0.4<br />
Figure 2: Hénon-like strange attractors <strong>of</strong> <strong>the</strong> model family (2) for (α, δ) <strong>in</strong> Arnol ′ d tongues <strong>of</strong><br />
periods two <strong>and</strong> three. (A) Parameters are fixed at a = 1.3, b = 0.3, ε = 0.2, (α, δ) = (0.51, 0.116).<br />
(B) Same as (A) for α = 0.33793.<br />
0<br />
y<br />
0.4<br />
w<br />
PSfrag replacements<br />
v<br />
u<br />
Figure 3: Quasi-periodic Hénon-like strange attractor <strong>of</strong> <strong>the</strong> model family (2). Parameter values<br />
are fixed at a = 1.85, b = −0.2, δ = 0, α = ( √ 5 − 1)/2, ε = 0.1. For a better visualisation <strong>of</strong> <strong>the</strong><br />
folds, <strong>the</strong> plot is given <strong>in</strong> <strong>the</strong> variables (u, v, w), where u = (r + 4) cos(θ), v = (r + 4) s<strong>in</strong>(θ), with<br />
r = x cos(θ) + 10y s<strong>in</strong>(θ), <strong>and</strong> w = −x s<strong>in</strong>(θ) + 10y cos(θ).<br />
be a dissipative (i.e., area contract<strong>in</strong>g) diffeomorphism, that is sufficiently smooth. Next,<br />
denote by R α : ¡ 1 → ¡ 1 <strong>the</strong> rigid rotation R α (θ) = θ + α. Then we def<strong>in</strong>e <strong>the</strong> family<br />
P α,ε :<br />
2 × ¡ 1 → 2 × ¡ 1 , (x, y, θ) ↦→ ( K 1 (x, y) + P 1 , K 2 (x, y) + P 2 , θ + α + P 3<br />
)<br />
, (6)<br />
<strong>of</strong> <strong>diffeomorphisms</strong>, where P j , for j = 1, . . . , 3, is a smooth function <strong>of</strong> (x, y, θ, α, ε) such<br />
that P j = 0 for ε = 0. Notice that <strong>the</strong> model (6) is not a skew product, but that <strong>the</strong>re<br />
is full coupl<strong>in</strong>g <strong>of</strong> <strong>the</strong> two constituents. A hyperbolic fixed po<strong>in</strong>t p <strong>of</strong> K (5) corresponds<br />
to a normally hyperbolic <strong>in</strong>variant circle C α,0 = {p} ¡ × 1 for <strong>the</strong> map P α,ε at ε = 0. By<br />
normal hyperbolicity <strong>the</strong> circle C α,0 is persistent under small perturbations [20, Theorem<br />
1.1]. Similar remarks go for <strong>the</strong> case where p is a hyperbolic periodic po<strong>in</strong>t. In <strong>the</strong> sequel we<br />
shall use this both for <strong>the</strong> case where p is a saddle <strong>and</strong> where p is a s<strong>in</strong>k.<br />
For numerical illustrations <strong>and</strong> discussion, a concrete version <strong>of</strong> (6) is used. It consists <strong>of</strong><br />
5
0.4<br />
0<br />
-0.4<br />
y<br />
rag replacements<br />
x<br />
-0.8<br />
θ<br />
(A)<br />
0 0.2 0.4 0.6 0.8 1<br />
0.4<br />
0<br />
rag replacements<br />
-0.4<br />
y<br />
-0.8<br />
θ<br />
(A)<br />
x<br />
-2 -1 0 1<br />
(B)<br />
Figure 4:(A) Quasi-periodic Hénon-like attractor <strong>of</strong> <strong>the</strong> model family (2), projection on <strong>the</strong> (θ, y)-<br />
plane. Parameter values are fixed at a = 0.8, b = 0.4, δ = 0, α = ( √ 5 − 1)/2, ε = 0.7, <strong>in</strong>itial<br />
conditions x 0 = 1.5, y 0 = 0, θ 0 = 0. (B) Same as (A), projection on <strong>the</strong> (x, y)-plane (<strong>in</strong> grey, <strong>in</strong> <strong>the</strong><br />
background). ‘Slices’ <strong>of</strong> <strong>the</strong> attractor for 2πθ ∈ [0.1 × j, 0.1 × j + 0.001], j = 0, 1, . . . , 62, are plotted<br />
<strong>in</strong> black.<br />
6
eplacements<br />
2<br />
1<br />
0<br />
-1<br />
x<br />
z<br />
-2<br />
(A)<br />
P (Σ)<br />
PSfrag replacements Σ<br />
0.8 1 1.2 1.4<br />
(A)<br />
Σ<br />
P (Σ)<br />
x<br />
z<br />
x<br />
ỹ<br />
1.938<br />
1.932<br />
1.926<br />
(B)<br />
1.24 1.3 1.36 1.42<br />
Figure 5: (A) Strange attractor A <strong>of</strong> <strong>the</strong> Po<strong>in</strong>caré return map <strong>of</strong> a climatological system [9].<br />
Compare with Figure 3. The attractor A is plotted with a ‘slice’ Σ <strong>and</strong> with <strong>the</strong> image <strong>of</strong> Σ under<br />
<strong>the</strong> return map. The slice Σ conta<strong>in</strong>s all po<strong>in</strong>ts with distance less that 0.0001 from <strong>the</strong> plane z = 0.<br />
The image <strong>of</strong> Σ is magnified <strong>in</strong> <strong>the</strong> central box. (B) Slice Σ <strong>of</strong> <strong>the</strong> attractor A <strong>in</strong> (A), projection<br />
on <strong>the</strong> (x, ỹ)-plane, with ỹ = y − 0.133 ∗ z.<br />
¡ ¡<br />
a perturbation <strong>of</strong> (2), where a coupl<strong>in</strong>g term <strong>in</strong> µy is added to <strong>the</strong> angle dynamics:<br />
⎛ ⎞<br />
x<br />
⎛<br />
⎞<br />
1 − (a + ε s<strong>in</strong>(2πθ))x 2 + y<br />
T = T α,δ,a,b,ε,µ :<br />
2 × 1 → 2 × 1 , ⎝y⎠ ↦→ ⎝<br />
θ<br />
bx<br />
θ + α + δ s<strong>in</strong>(2πθ) + µy<br />
⎠ , (7)<br />
depend<strong>in</strong>g on <strong>the</strong> six parameters (α, δ, a, b, ε, µ).<br />
1.2 Motivation<br />
Quasi-periodic Hénon-like attractors have been conjectured to occur <strong>in</strong> <strong>diffeomorphisms</strong> <strong>of</strong><br />
3 = {x, y, z}, obta<strong>in</strong>ed as Po<strong>in</strong>caré return maps for a climatological model [9, 10, 41],<br />
compare <strong>the</strong> attractor A displayed <strong>in</strong> Figure 5. Exam<strong>in</strong>ation <strong>of</strong> a cross-section Σ <strong>of</strong> <strong>the</strong><br />
attractor (magnified <strong>in</strong> Figure 5 (B)) suggests that A is conta<strong>in</strong>ed <strong>in</strong> a two-dimensional<br />
manifold which is folded onto itself, <strong>in</strong> analogy with <strong>the</strong> structure <strong>of</strong> <strong>the</strong> Hénon attractor [18].<br />
This manifold supposedly is <strong>the</strong> unstable manifold W u (C ) <strong>of</strong> a <strong>quasi</strong>-periodic <strong>in</strong>variant circle<br />
C <strong>of</strong> saddle type. To illustrate <strong>the</strong> dynamics <strong>in</strong>side A we computed <strong>the</strong> image <strong>of</strong> <strong>the</strong> slice Σ<br />
under <strong>the</strong> return map. This yields a folded curve look<strong>in</strong>g like a planar Hénon attractor.<br />
Remark 2. Also we mention that <strong>the</strong> occurrence <strong>of</strong> strange attractors which look similar<br />
to Figure 4 (A) is observed <strong>in</strong> [28, 15, 17, 22, 29]. Although most <strong>of</strong> <strong>the</strong>se studies deal with<br />
endomorphisms <strong>of</strong> <strong>the</strong> <strong>in</strong>terval forced by a rigid rotation <strong>in</strong> a skew product way, <strong>and</strong> some <strong>of</strong><br />
<strong>the</strong>m have negative Lyapunov exponents (beyond <strong>the</strong> one trivially equal to zero), <strong>the</strong>re may<br />
be a relationship with <strong>the</strong> present approach. See Sections 2.3 <strong>and</strong> 4 for fur<strong>the</strong>r discussion.<br />
The <strong>the</strong>oretical knowledge <strong>of</strong> attractors <strong>in</strong> higher dimension is limited. As positive exceptions<br />
to this we mention Viana [39, 40], Tatjer [36] <strong>and</strong> Wang & Young [42]. The Hénon-like<br />
attractors found <strong>in</strong> <strong>the</strong> present paper, to some extent, also belong to this doma<strong>in</strong>. In this<br />
sense one may say that <strong>the</strong> underst<strong>and</strong><strong>in</strong>g <strong>of</strong> <strong>the</strong> <strong>quasi</strong>-periodic Hénon-like attractor is a<br />
next step <strong>in</strong> this research program. A more detailed discussion <strong>and</strong> fur<strong>the</strong>r motivation <strong>of</strong> <strong>the</strong><br />
search for <strong>quasi</strong>-periodic Hénon-like attractors is postponed to Section 2.3.<br />
7
1.3 Summary <strong>and</strong> outl<strong>in</strong>e<br />
We summarise <strong>the</strong> rema<strong>in</strong>der <strong>of</strong> this paper, also expla<strong>in</strong><strong>in</strong>g its organisation. First <strong>in</strong> Section 2<br />
<strong>the</strong> results are presented, where all longer pro<strong>of</strong>s are postponed to Section 3. The contents <strong>of</strong><br />
Section 2 are related as follows to <strong>the</strong> subdivision regard<strong>in</strong>g model (2) as given <strong>in</strong> Section 1.1.<br />
Numerical examples beyond <strong>the</strong> skew product model (2), as well as details on methods <strong>and</strong><br />
on <strong>the</strong> <strong>in</strong>terpretation <strong>of</strong> <strong>the</strong> results, are given <strong>in</strong> Section 4.<br />
Normally hyperbolic <strong>in</strong>variant circles<br />
In Section 2.1 we start consider<strong>in</strong>g <strong>the</strong> more general context <strong>of</strong> <strong>the</strong> fully coupled family (6).<br />
A hyperbolic periodic po<strong>in</strong>t p, for ε = 0 corresponds to a normally hyperbolic <strong>in</strong>variant circle<br />
C .<br />
We start with <strong>the</strong> case where p is a saddle-po<strong>in</strong>t, so where C normally is <strong>of</strong> saddle type as<br />
well. Theorem 2 asserts that now, under quite general conditions, <strong>the</strong> closure Cl (W u (C )) <strong>of</strong><br />
<strong>the</strong> unstable manifold <strong>of</strong> C attracts an open set. In <strong>the</strong> families (6) <strong>and</strong> (2) this corresponds<br />
to an open set <strong>of</strong> parameter values. Note that <strong>the</strong> attract<strong>in</strong>g set Cl (W u (C )) is not m<strong>in</strong>imal<br />
if C carries Morse-Smale dynamics. In this situation, <strong>the</strong> case 2(a) <strong>of</strong> Section 2 is partially<br />
covered.<br />
Next, if p is a hyperbolic periodic po<strong>in</strong>t <strong>of</strong> <strong>the</strong> family (6), Theorem 3 guarantees that C<br />
is <strong>quasi</strong>-periodic for a parameter set <strong>of</strong> positive measure. If p is a saddle-po<strong>in</strong>t, <strong>in</strong> comb<strong>in</strong>ation<br />
with <strong>the</strong> previous paragraph, we partially cover case 2(b) <strong>of</strong> Section 1.1 for <strong>the</strong> family<br />
(2). However, if p is a periodic s<strong>in</strong>k, it follows that C is a <strong>quasi</strong>-periodic attractor, which<br />
completely covers case 1(b) <strong>of</strong> Section 1.1.<br />
In <strong>the</strong> case where p is a s<strong>in</strong>k <strong>and</strong> C carries Morse-Smale dynamics, for <strong>the</strong> general sett<strong>in</strong>g<br />
<strong>of</strong> (6), <strong>the</strong> m<strong>in</strong>imal attractors are periodic. For <strong>the</strong> family (2) this completely covers case<br />
1(a) <strong>of</strong> Section 1.1.<br />
Persistence <strong>of</strong> normally hyperbolic <strong>in</strong>variant circles generally follows by [20, Theorem 1.1].<br />
Theorem 2 is based on a result by Tangerman <strong>and</strong> Szewc, see [31, Appendix 3]. Our pro<strong>of</strong><br />
<strong>of</strong> Theorem 3 applies st<strong>and</strong>ard kam Theory, see [6, 5].<br />
The cases where Λ H > 0<br />
We now deal with <strong>the</strong> cases where <strong>the</strong> Hénon family is <strong>in</strong> a strange attractor. The ma<strong>in</strong><br />
ma<strong>the</strong>matical contents <strong>of</strong> this paper are formulated <strong>in</strong> Theorem 4 (<strong>in</strong> Section 2.2), which<br />
deals with <strong>the</strong> scaled skew product model (4). Here we establish <strong>the</strong> existence <strong>of</strong> Hénon-like<br />
strange attractors, which completely covers case 2(a) <strong>of</strong> Section 1.1 <strong>in</strong> suitable parameter<br />
ranges. For completeness, <strong>in</strong> Section 2.2 technical def<strong>in</strong>itions <strong>of</strong> ‘strange’, ‘Hénon-like’, etc.,<br />
are <strong>in</strong>cluded; compare with [14, 26, 39]. The pro<strong>of</strong> <strong>of</strong> Theorem 4 is based on [14, Theorem<br />
5.2].<br />
F<strong>in</strong>ally, <strong>in</strong> Section 2.3 we deal with <strong>the</strong> rema<strong>in</strong><strong>in</strong>g case, both for <strong>the</strong> skew product system<br />
(4) <strong>and</strong> for <strong>the</strong> more general system (6). Here we touch upon <strong>the</strong> existence <strong>of</strong> <strong>quasi</strong>-periodic<br />
Hénon-like attractors. Lemma 7 implies that <strong>in</strong> <strong>the</strong> product case ε = 0 a topologically<br />
transitive attractor occurs, hav<strong>in</strong>g a dense set <strong>of</strong> orbits with a positive Lyapunov exponent.<br />
Regard<strong>in</strong>g its persistence for |ε| ≪ 1, our conjectures <strong>and</strong> mostly computer suggested.<br />
Acknowledgements<br />
The authors are <strong>in</strong>debted to Henk Bru<strong>in</strong>, Àngel Jorba, Marco Martens, V<strong>in</strong>cent Naudot,<br />
Floris Takens, Joan Carles Tatjer <strong>and</strong> Marcelo Viana for valuable discussion. The first author<br />
8
is <strong>in</strong>debted to <strong>the</strong> Departament de Matemàtica Aplicada i Anàlisi, Universitat de Barcelona,<br />
for hospitality <strong>and</strong> <strong>the</strong> last two authors are <strong>in</strong>debted to <strong>the</strong> Department <strong>of</strong> Ma<strong>the</strong>matics,<br />
University <strong>of</strong> Gron<strong>in</strong>gen, for <strong>the</strong> same reason. The research <strong>of</strong> C.S. has been supported by<br />
grants DGICYT BFM2003-09504-C02-01 (Spa<strong>in</strong>) <strong>and</strong> CIRIT 2001 SGR-70 (Catalonia). The<br />
comput<strong>in</strong>g cluster HIDRA <strong>of</strong> <strong>the</strong> UB Group <strong>of</strong> Dynamical Systems have been widely used.<br />
We are <strong>in</strong>debted to J. Timoneda for keep<strong>in</strong>g it fully operative.<br />
2 Statement <strong>of</strong> <strong>the</strong> results<br />
As announced before, our results are formulated <strong>in</strong> <strong>the</strong> next subsections, while pro<strong>of</strong>s are<br />
given <strong>in</strong> Sections 3.1 <strong>and</strong> 3.2.<br />
2.1 Invariant circles <strong>of</strong> saddle-type <strong>and</strong> bas<strong>in</strong>s <strong>of</strong> attraction<br />
We first consider maps F <strong>of</strong> <strong>the</strong> <strong>solid</strong> <strong>torus</strong><br />
2 × ¡ 1 obta<strong>in</strong>ed by perturb<strong>in</strong>g <strong>the</strong> product <strong>of</strong> a<br />
¡<br />
¡ ¡<br />
planar map times a rotation on 1 . Assum<strong>in</strong>g that <strong>the</strong> planar map has a saddle fixed po<strong>in</strong>t<br />
with a transversal homocl<strong>in</strong>ic po<strong>in</strong>t, it is proved that <strong>the</strong> map F has an attractor conta<strong>in</strong>ed<br />
<strong>in</strong>side Cl (W u (C )).<br />
To this end we generalise an unpublished result <strong>of</strong> Tangerman <strong>and</strong> Szewc, see [31, Appendix<br />
3], where we are <strong>in</strong> <strong>the</strong> general context <strong>of</strong> <strong>the</strong> family P α,ε : 2 × 1 → 2 × 1 , see<br />
(6).<br />
Proposition 1. (normally hyperbolic <strong>in</strong>variant circle) Suppose that K has a hyperbolic<br />
fixed po<strong>in</strong>t p = (x 0 , y 0 ). Then for all α ∈ [0, 1] <strong>the</strong> map P α,0 has a normally hyperbolic<br />
<strong>in</strong>variant circle C α,0 = {p} × ¡ 1 . The manifold C α,0 is r-normally hyperbolic for all <strong>in</strong>tegers<br />
r with 1 ≤ r ≤ n. Moreover, for all r < n <strong>the</strong>re exists an ε r > 0 such that for all ε < ε r<br />
<strong>and</strong> all α ∈ [0, 1], P α,ε has a normally hyperbolic <strong>in</strong>variant circle C α,ε <strong>of</strong> class C r , which is<br />
C r -close to C α,0 .<br />
Pro<strong>of</strong>: The dynamics <strong>of</strong> P α,0 on C α,0 is parallel with rotation number α. This implies that<br />
C α,0 is an r-normally hyperbolic <strong>in</strong>variant manifold for all r ≤ n <strong>and</strong>, <strong>the</strong>refore, it is <strong>of</strong><br />
class C n . So C α,0 (as well as its stable <strong>and</strong> unstable manifolds), is persistent under C n -small<br />
perturbations. This directly follows from [20, Theorem 1.1].<br />
Proposition 1 allows us to construct a bas<strong>in</strong> <strong>of</strong> attraction with nonempty <strong>in</strong>terior for <strong>the</strong><br />
<strong>in</strong>variant set Cl (W u (C α,ε )), provided that p is a saddle po<strong>in</strong>t, while <strong>the</strong> one-dimensional<br />
unstable manifold W u (p) ⊂ 2 <strong>of</strong> <strong>the</strong> map K, see (5), does not escape to <strong>in</strong>f<strong>in</strong>ity. For<br />
(x, y, θ) ∈ 2 × ¡ 1 , denote by ω(x, y, θ) <strong>the</strong> ω-limit set <strong>of</strong> (x, y, θ) under P α,ε .<br />
Theorem 2. (attractor conta<strong>in</strong>ed <strong>in</strong> Cl (W u (C ))) Fix <strong>in</strong>tegers n <strong>and</strong> r such that<br />
n ≥ 2 <strong>and</strong> 1 ≤ r < n. Choose ε < ε r as <strong>in</strong> Proposition 1 <strong>and</strong> let α ∈ [0, 1]. Suppose that<br />
K : 2 → 2 is <strong>of</strong> class C n <strong>and</strong> satisfies:<br />
1. K has a saddle fixed po<strong>in</strong>t p ∈ 2 <strong>and</strong> a transversal homocl<strong>in</strong>ic po<strong>in</strong>t q ∈ W s (p)∩W u (p).<br />
2. K is uniformly dissipative: <strong>the</strong>re exists κ < 1 such that |det(DK(x, y))| ≤ κ for all<br />
(x, y) ∈ 2 .<br />
3. W u (p) is conta<strong>in</strong>ed <strong>in</strong> a bounded subset <strong>of</strong> 2 .<br />
9
Then <strong>the</strong>re exists an ε ∗ < ε r such that for all ε < ε ∗ <strong>the</strong>re exists an open, nonempty bounded<br />
set U ⊂<br />
2 × ¡ 1 such that for all (x, y, θ) ∈ U<br />
ω(x, y, θ) ⊂ Cl (W u (C α,ε )) . (8)<br />
Remark 3. By tak<strong>in</strong>g iterates <strong>of</strong> <strong>the</strong> map P α,ε , Theorem 2 can be adapted to <strong>the</strong> case<br />
where p is a saddle periodic po<strong>in</strong>t. In this context we have <strong>the</strong> <strong>in</strong>clusion (8), where C α,ε is a<br />
periodically <strong>in</strong>variant circle, i.e. a circle which is <strong>in</strong>variant under some iterate <strong>of</strong> P α,ε .<br />
Under <strong>the</strong> conditions <strong>of</strong> Theorem 2, <strong>the</strong> <strong>in</strong>variant set Cl (W u (C α,ε )) attracts all orbits with<br />
<strong>in</strong>itial state <strong>in</strong> an open set U. This holds for an open set <strong>of</strong> ε-values. In general, however,<br />
Cl (W u (C α,ε )) is not an attractor, s<strong>in</strong>ce it might be non-topologically transitive. This occurs,<br />
for example, if Cl (W u (C α,ε )) conta<strong>in</strong>s a periodic attractor.<br />
In <strong>the</strong> next Theorem we prove that at least <strong>the</strong> circle C α,ε is <strong>quasi</strong>-periodic (<strong>and</strong>, hence,<br />
topologically transitive) for a set <strong>of</strong> parameter values hav<strong>in</strong>g large relative measure.<br />
Theorem 3. (normally hyperbolic <strong>quasi</strong>-periodic circles) Let P α,ε be a C n -family<br />
<strong>of</strong> <strong>diffeomorphisms</strong> as <strong>in</strong> (6), where n ≥ 5. Choose ε r as <strong>in</strong> Proposition 1. Then <strong>the</strong>re exists<br />
an ε ∗∗ < ε r such that for all ε < ε ∗∗ <strong>the</strong> follow<strong>in</strong>g holds.<br />
1. There exists a set D ε ⊂ [0, 1] with Lebesgue measure meas(D ε ) > 0 such that for<br />
α ∈ D ε <strong>the</strong> restriction <strong>of</strong> P α,ε to <strong>the</strong> circle C α,ε is smoothly conjugate to an irrational<br />
rigid rotation.<br />
2. meas(D ε ) tends to 1 for ε → 0.<br />
Pro<strong>of</strong>s <strong>of</strong> Theorems 2 <strong>and</strong> 3 are given <strong>in</strong> Section 3.1.<br />
Theorem 3, as happens with Theorem 2, has a direct analogue for <strong>the</strong> case where p is<br />
a hyperbolic periodic po<strong>in</strong>t. The Theorems will be applied both for <strong>the</strong> case where p is<br />
a periodic saddle <strong>and</strong> a periodic s<strong>in</strong>k. In <strong>the</strong> latter case we prove <strong>the</strong> existence <strong>of</strong> <strong>quasi</strong>periodic<br />
attractors for positive measure <strong>in</strong> parameter space, as described <strong>in</strong> <strong>the</strong> case 2(b) <strong>of</strong><br />
Section 1.1. Aga<strong>in</strong> see Figure 1 (C). This situation corresponds to po<strong>in</strong>ts with code 2, <strong>in</strong><br />
blue.<br />
A complementary situation regard<strong>in</strong>g Theorem 3 occurs when <strong>the</strong> dynamics on C α,ε is<br />
<strong>of</strong> Morse-Smale type, compare with <strong>the</strong> resonance tongues <strong>of</strong> Figure 1 (A). In that case,<br />
for Λ H < 0, <strong>the</strong> attract<strong>in</strong>g set Cl (W u (C α,ε )) <strong>of</strong> system (6) conta<strong>in</strong>s a hyperbolic periodic<br />
attractor as described <strong>in</strong> case 1(a) <strong>of</strong> Section 1.1. Aga<strong>in</strong> compare with Figure 1 (C), po<strong>in</strong>ts<br />
with code 1, <strong>in</strong> yellow. In <strong>the</strong> next section, for <strong>the</strong> skew product system (4) we show that <strong>in</strong><br />
<strong>the</strong> case where Λ H > 0, <strong>the</strong> set Cl (W u (C α,ε )) conta<strong>in</strong>s a Hénon-like attractor, which covers<br />
<strong>the</strong> case 2(a) <strong>of</strong> Section 1.1, compare with Figure 1 (C), po<strong>in</strong>ts with code 3, <strong>in</strong> red.<br />
2.2 Hénon-like attractors <strong>in</strong> a family <strong>of</strong> skew product maps<br />
By Theorem 2, <strong>the</strong> set Cl (W u (C )) is attract<strong>in</strong>g under quite general circumstances. As may<br />
be clear from <strong>the</strong> previous paragraph, <strong>in</strong> general Cl (W u (C )) does not have to be topologically<br />
transitive, <strong>in</strong> which case it is not considered an attractor. (For precise def<strong>in</strong>itions see below.)<br />
However, <strong>in</strong> <strong>the</strong> particular case <strong>of</strong> map (2), we show that Cl (W u (C )) conta<strong>in</strong>s Hénon-like<br />
attractors. We first recall a few basic def<strong>in</strong>itions regard<strong>in</strong>g strange attractors, suited to our<br />
purposes.<br />
Def<strong>in</strong>ition 1. [14, 26, 39] Let F : M → M be a C 1 -diffeomorphism, where M is an<br />
m-dimensional smooth manifold.<br />
10
1. An F -<strong>in</strong>variant set A ⊂ M is called topologically transitive if <strong>the</strong>re exists a po<strong>in</strong>t<br />
z ∈ A such that <strong>the</strong> orbit Orb(z) = {F j (z)} j≥0 <strong>of</strong> z under F is dense <strong>in</strong> A .<br />
2. A set A ⊂ M is called an attractor if it is topologically transitive, compact, F -<strong>in</strong>variant<br />
<strong>and</strong> if <strong>the</strong> stable set (bas<strong>in</strong> <strong>of</strong> attraction) W s (A ) has nonempty <strong>in</strong>terior.<br />
3. An attractor A is called strange if <strong>the</strong>re exist constants κ > 0, λ > 1, a dense orbit<br />
Orb(z) ⊂ A <strong>and</strong> a vector v ∈ T z M such that<br />
‖DF n (z)v‖ ≥ κλ n for n ≥ 0.<br />
4. The attractor A is called Hénon-like if <strong>the</strong>re exist a saddle periodic orbit Orb(p) =<br />
{p, F (p), . . . , F n (p)}, a po<strong>in</strong>t z <strong>in</strong> <strong>the</strong> unstable manifold W u (Orb(p)), constants κ > 0,<br />
λ > 1, <strong>and</strong> tangent vectors v, w ∈ T z M, with w ≠ 0, such that<br />
where Cl(·) denotes topological closure.<br />
i) A = Cl (W u (Orb(p))) , (9)<br />
ii) Orb(z) is dense <strong>in</strong> A , (10)<br />
iii) ‖DF n (z)v‖ ≥ κλ n for n ≥ 0, (11)<br />
iv) ‖DF n (z)w‖ → 0 as n → ±∞, (12)<br />
In particular, Hénon-like attractors are strange, s<strong>in</strong>ce by conditions (10) <strong>and</strong> (11) <strong>the</strong>y admit<br />
a dense orbit with a positive Lyapunov exponent. Moreover, Hénon-like attractors are nonuniformly<br />
hyperbolic; <strong>in</strong>deed, by condition (12) <strong>the</strong>y conta<strong>in</strong> critical po<strong>in</strong>ts, that is, po<strong>in</strong>ts<br />
belong<strong>in</strong>g to a dense orbit for which a nonzero tangent vector w exists, which is contracted<br />
both by positive <strong>and</strong> by negative iteration <strong>of</strong> <strong>the</strong> derivative DF .<br />
We now come to <strong>the</strong> ma<strong>in</strong> result <strong>of</strong> <strong>the</strong> present paper, regard<strong>in</strong>g <strong>the</strong> occurrence <strong>of</strong> Hénonlike<br />
strange attractors <strong>in</strong> <strong>the</strong> scaled skew product family (4). First we recall that <strong>the</strong> restriction<br />
<strong>of</strong> (4) to ¡ 1 is <strong>the</strong> Arnol ′ d family <strong>of</strong> circle maps (3). Moreover, <strong>the</strong> map (4) is a<br />
generalisation <strong>of</strong> <strong>the</strong> planar Hénon-like families considered <strong>in</strong> [26, 39]. The latter are families<br />
<strong>of</strong> planar <strong>diffeomorphisms</strong>, which are C 3 -small perturbations <strong>of</strong> <strong>the</strong> Logistic family<br />
Q a : → , x ↦→ 1 − ax 2 . (13)<br />
The x- <strong>and</strong> y-components <strong>of</strong> T α,δ,a,ε also depend on <strong>the</strong> circle dynamics by <strong>the</strong> perturbative<br />
terms f <strong>and</strong> g. The only requirement on f <strong>and</strong> g is that <strong>the</strong>ir C 3 -norms are bounded on<br />
compact sets. Occurrence <strong>of</strong> Hénon-like attractors is proved <strong>in</strong> <strong>the</strong> family T α,δ,a,ε for all<br />
parameter values belong<strong>in</strong>g to a set <strong>of</strong> positive (Lebesgue) measure. For all values <strong>in</strong> this<br />
set, <strong>the</strong> parameters (α, δ) are such that <strong>the</strong> dynamics <strong>of</strong> <strong>the</strong> Arnol ′ d family A α,δ (3) is <strong>of</strong><br />
Morse-Smale type: <strong>the</strong>re exist periodic po<strong>in</strong>ts θ s <strong>and</strong> θ r ¡ <strong>in</strong> 1 , such that θ s is attract<strong>in</strong>g <strong>and</strong><br />
θ r repell<strong>in</strong>g for A α,δ . By A q/n we denote <strong>the</strong> open resonance tongue <strong>in</strong> <strong>the</strong> (α, δ)-plane where<br />
<strong>the</strong>se periodic po<strong>in</strong>ts have rotation number q/n [1, 13] <strong>and</strong> <strong>the</strong> width <strong>of</strong> <strong>the</strong> tongue <strong>in</strong> α<br />
behaves as δ n [8], compare with Figure 1 (A). The parameter space under consideration is<br />
<strong>the</strong> set <strong>of</strong> all (α, δ, a, ε) ∈<br />
4 such that<br />
α ∈ [0, 1], δ ∈ [ 0, 1/(2π) ) , a ∈ [0, 2], |ε| < 1. (14)<br />
The attractors A we obta<strong>in</strong>, co<strong>in</strong>cide with <strong>the</strong> closure <strong>of</strong> <strong>the</strong> one-dimensional unstable manifold<br />
A = Cl (W u (Orb(p))) ,<br />
where p = (x 0 , y 0 , θ s ) ∈<br />
2 ¡ × 1 belongs to a hyperbolic periodic orbit <strong>of</strong> saddle type. For<br />
<strong>the</strong> statement <strong>of</strong> <strong>the</strong> result we need a few def<strong>in</strong>itions <strong>and</strong> notations.<br />
11
Def<strong>in</strong>ition 2. 1. A map M : J → J, where J ⊂ is an <strong>in</strong>terval, is called topologically<br />
mix<strong>in</strong>g if for any open <strong>in</strong>tervals J 1 , J 2 ⊂ J <strong>the</strong>re exists n 0 such that<br />
M n (J 1 ) ∩ J 2 ≠ ∅ for all n ≥ n 0 .<br />
2. The <strong>in</strong>terval K a = [Q 2 a (0), Q a(0)] is called <strong>the</strong> core or <strong>the</strong> restrictive <strong>in</strong>terval <strong>of</strong> <strong>the</strong><br />
Logistic family Q a (13).<br />
It is well-known that Q a ([0, 1]) = Q a (K a ) = K a for all a, where K a is <strong>the</strong> core <strong>of</strong> Q a (13),<br />
see e.g. [24, Section II.5]. For a given <strong>in</strong>teger n > 1, denote by Φ(n) <strong>the</strong> set <strong>of</strong> all <strong>in</strong>tegers q<br />
such that q <strong>and</strong> n are relatively prime, where 1 ≤ q < n. For n = 1 we put Φ(n) = {1}.<br />
Theorem 4. (Hénon-like attractors <strong>in</strong> (4)) Choose a ∗ ∈ (1, 2) such that <strong>the</strong> quadratic<br />
map Q a ∗ <strong>in</strong> (13) is topologically mix<strong>in</strong>g on its core K = [1 − a ∗ , 1] <strong>and</strong> its critical po<strong>in</strong>t c = 0<br />
is preperiodic. Let n ≥ 1 be an <strong>in</strong>teger. There exist a periodic po<strong>in</strong>t p 0 <strong>of</strong> <strong>the</strong> n-th iterate Q n a ∗<br />
<strong>and</strong> positive constants ¯ε n , ā n <strong>and</strong> χ n such that <strong>the</strong> follow<strong>in</strong>g holds.<br />
1. For all (α, δ, a, ε) as <strong>in</strong> (14), with<br />
(α, δ) ∈ ∪ q∈Φ(n) Cl ( A q/n) , |a − a ∗ | < ā n , |ε| < ¯ε n (15)<br />
<strong>the</strong> map T α,δ,a,ε has a saddle periodic po<strong>in</strong>t p, which is <strong>the</strong> analytic cont<strong>in</strong>uation <strong>of</strong> p 0<br />
<strong>and</strong> such that <strong>the</strong> unstable manifold W u (Orb(p)) is one-dimensional.<br />
2. For all (α, δ, ε) as <strong>in</strong> (15) <strong>the</strong>re exists a set S α,δ,ε with<br />
S α,δ,ε ⊂ [a ∗ − ā n , a ∗ + ā n ],<br />
meas(S) > χ n<br />
such that for all a ∈ S α,δ,ε <strong>the</strong> closure Cl (W u (Orb(p))) is a Hénon-like attractor <strong>of</strong><br />
T α,δ,a,ε .<br />
Corollary 5. The set <strong>of</strong> parameter values for which T α,δ,a,ε has a Hénon-like attractor conta<strong>in</strong>s<br />
<strong>the</strong> set<br />
S = ⋃ n∈<br />
{<br />
(α, δ, a, ε) | (α, δ) ∈ ∪q∈Φ(n) Cl ( A q/n) , |ε| < ¯ε n , a ∈ S α,δ,ε<br />
}<br />
,<br />
<strong>and</strong> <strong>the</strong> set S has positive Lebesgue measure<br />
meas(S) ≥ 2<br />
∞∑<br />
¯ε n χ n<br />
n=1<br />
∑<br />
q∈Φ(n)<br />
meas A q/n .<br />
Our pro<strong>of</strong> <strong>of</strong> Theorem 4 is given <strong>in</strong> Section 3.2. It is based on a result <strong>of</strong> Díaz-Rocha-Viana [14,<br />
Theorem 5.2], <strong>and</strong> relies on <strong>the</strong> follow<strong>in</strong>g facts:<br />
1. For (α, δ) <strong>in</strong>side any tongue A q/n , <strong>the</strong> asymptotic dynamics <strong>of</strong> T α,δ,a,ε is described by<br />
an O(ε)-perturbation <strong>of</strong> <strong>the</strong> n-th iterate Q n a .<br />
2. For all n <strong>the</strong> map Q n a is a generic n-modal family, <strong>in</strong> <strong>the</strong> sense <strong>of</strong> [14, Section 5.2],<br />
also see <strong>the</strong> def<strong>in</strong>ition given <strong>in</strong> Section 3.2. To show this, we use that Q a ∗ is a Misiurewicz<br />
map [25], <strong>and</strong>, <strong>the</strong>refore, it is Collet-Eckmann (see e.g. [24, Section V.4]). See<br />
Section 3.2 for details.<br />
12
Notice that <strong>the</strong> family (2) takes <strong>the</strong> form (4) after a rescal<strong>in</strong>g y ↦→ √ |b|y <strong>and</strong> by choos<strong>in</strong>g b =<br />
O(ε). Therefore, by restrict<strong>in</strong>g <strong>the</strong> parameter δ to sufficiently small values, both Theorem 4<br />
<strong>and</strong> Theorem 2 may be applied to (2).<br />
Corollary 6. Let a ∗ <strong>and</strong> p 0 satisfy <strong>the</strong> hypo<strong>the</strong>ses <strong>of</strong> Theorem 4. Then <strong>the</strong>re exists a positive<br />
measure set <strong>of</strong> parameter values such that <strong>the</strong> family (2) has Hénon-like attractors, conta<strong>in</strong>ed<br />
<strong>in</strong> <strong>the</strong> closure <strong>of</strong> <strong>the</strong> unstable manifold <strong>of</strong> a periodically <strong>in</strong>variant circle.<br />
Pro<strong>of</strong>: Take a ∗ <strong>and</strong> p 0 as <strong>in</strong> <strong>the</strong> hypo<strong>the</strong>ses <strong>of</strong> Theorem 4. Then for all δ <strong>and</strong> for ε <strong>and</strong> b<br />
sufficiently small, Theorem 4 applies. Moreover, for (ε, δ) = (0, 0) <strong>the</strong> circle C = {p 0 } × ¡ 1 is<br />
periodically <strong>in</strong>variant under map (2). In particular, <strong>the</strong> conditions <strong>of</strong> Theorem 2 are satisfied<br />
for b sufficiently small, s<strong>in</strong>ce:<br />
1. The periodic po<strong>in</strong>t (p 0 , 0) <strong>of</strong> H a,0 has an analytic cont<strong>in</strong>uation ¯p(b) for all b sufficiently<br />
small, <strong>and</strong> p 0 is chosen such that ¯p(b) has transversal homocl<strong>in</strong>ic po<strong>in</strong>ts, see Proposition<br />
12.<br />
2. det(DH a,b (x, y)) = b;<br />
3. The unstable manifold <strong>of</strong> all periodic po<strong>in</strong>ts <strong>of</strong> H a,b is bounded for b sufficiently small,<br />
s<strong>in</strong>ce <strong>the</strong> <strong>in</strong>variant manifolds depend cont<strong>in</strong>uously on <strong>the</strong> map [26, Prop. 7.1].<br />
So for (ε, δ, b) sufficiently small, <strong>the</strong> conclusions <strong>of</strong> Theorem 2 hold.<br />
Two attractors occurr<strong>in</strong>g <strong>in</strong> <strong>the</strong> family (2) are shown <strong>in</strong> Figure 4 (A) <strong>and</strong> (B), for (α, δ) <strong>in</strong> an<br />
Arnol ′ d resonance tongue <strong>of</strong> period two <strong>and</strong> three, respectively. Also compare with Figure<br />
1 (C), po<strong>in</strong>ts with code 3, <strong>in</strong> red. It is to be noted that <strong>the</strong> Hénon-like character <strong>of</strong> <strong>the</strong>se<br />
attractors for larger values <strong>of</strong> b <strong>and</strong> ε rema<strong>in</strong>s conjectural.<br />
2.3 Quasi-periodic Hénon-like attractors<br />
We start with a fur<strong>the</strong>r sett<strong>in</strong>g <strong>of</strong> <strong>the</strong> problems regard<strong>in</strong>g <strong>quasi</strong>-periodic Hénon-like attractors,<br />
<strong>in</strong> <strong>the</strong> skew product model family (2).<br />
2.3.1 Fur<strong>the</strong>r sett<strong>in</strong>g <strong>of</strong> <strong>the</strong> problem<br />
The present paper has been partially motivated by <strong>the</strong> problem to f<strong>in</strong>d a diffeomorphism F<br />
with a strange attractor A such that<br />
A = Cl (W u (C )) , (16)<br />
where C is an F -<strong>in</strong>variant circle <strong>of</strong> saddle type with irrational rotation number, so with <strong>quasi</strong>periodic<br />
dynamics. In this context, <strong>the</strong> role <strong>of</strong> <strong>the</strong> saddle periodic orbit <strong>in</strong> (9) is played by<br />
a <strong>quasi</strong>-periodic <strong>in</strong>variant circle <strong>of</strong> saddle type. By analogy with <strong>the</strong> def<strong>in</strong>ition <strong>of</strong> Hénon-like<br />
strange attractors (see Section 2.2), we are led to <strong>the</strong> follow<strong>in</strong>g def<strong>in</strong>ition.<br />
Def<strong>in</strong>ition 3. Let F : M → M be a C 1 -diffeomorphism, where M is an m-dimensional<br />
smooth manifold. We say that <strong>the</strong> attractor A is <strong>quasi</strong>-periodic Hénon-like if <strong>the</strong>re exist<br />
1. A <strong>quasi</strong>-periodic <strong>in</strong>variant circle C <strong>of</strong> saddle type such that A = Cl (W u (C )) .<br />
2. A po<strong>in</strong>t x ∈ A such that Orb(x) is dense <strong>in</strong> A <strong>and</strong><br />
3. a dense set Z ⊂ A <strong>and</strong> constants κ > 0, λ > 1 such that for all z ∈ Z <strong>the</strong>re exist<br />
vectors v, w ∈ T z M such that conditions (11) <strong>and</strong> (12) hold.<br />
13
The def<strong>in</strong>ition mimics <strong>the</strong> positive Lyapunov exponents <strong>and</strong> non-uniform hyperbolicity requirements<br />
<strong>in</strong> <strong>the</strong> def<strong>in</strong>ition <strong>of</strong> Hénon-like attractors <strong>and</strong> also asks for transitivity. As usual<br />
similar def<strong>in</strong>itions can be given with F replaced by a power F k .<br />
Return<strong>in</strong>g to <strong>the</strong> skew product context <strong>of</strong> <strong>the</strong> model family (2), <strong>in</strong> <strong>the</strong> Arnol ′ d family A α,δ<br />
we fix parameter values (α, δ) such that <strong>the</strong> dynamics <strong>of</strong> A α,δ is <strong>quasi</strong>-periodic. Recall that<br />
<strong>the</strong> set <strong>of</strong> all such (α, δ) has positive measure <strong>and</strong> is nowhere dense [5, Chap. 1]. Next choose<br />
parameter values a <strong>and</strong> b such that <strong>the</strong> Hénon map (1) has a Hénon-like strange attractor<br />
A ′ , co<strong>in</strong>cid<strong>in</strong>g with <strong>the</strong> closure <strong>of</strong> <strong>the</strong> unstable manifold <strong>of</strong> a saddle fixed po<strong>in</strong>t p. Also recall<br />
that, accord<strong>in</strong>g to [2, 3, 26], such (a, b) form a set <strong>of</strong> positive measure. Then, at ε = 0 <strong>the</strong><br />
map (2) has an attractor A = A ′ × ¡ 1 co<strong>in</strong>cid<strong>in</strong>g with <strong>the</strong> closure <strong>of</strong> <strong>the</strong> unstable manifold<br />
<strong>of</strong> <strong>the</strong> <strong>quasi</strong>-periodic saddle-type <strong>in</strong>variant circle {p}ס 1 . It may be clear that requirement 3<br />
<strong>of</strong> Def<strong>in</strong>ition 3 is satisfied by tak<strong>in</strong>g Z = Orb(z) × ¡ 1 , where z is a po<strong>in</strong>t satisfy<strong>in</strong>g properties<br />
4 ii), iii) <strong>and</strong> iv) <strong>in</strong> Def<strong>in</strong>ition 1 <strong>of</strong> Hénon-like attractors.<br />
Next, to prove that <strong>in</strong> <strong>the</strong> product case we obta<strong>in</strong> a <strong>quasi</strong>-periodic Hénon-like attractors<br />
only item 2 <strong>in</strong> Def<strong>in</strong>ition 3 has to be verified. This is done <strong>in</strong> <strong>the</strong> follow<strong>in</strong>g lemma.<br />
Lemma 7. (Transitivity <strong>of</strong> (2), uncoupled) Let T be a dissipative C 1 -diffeomorphism<br />
<strong>in</strong> an open subset U ⊂<br />
2 such that<br />
1. T has a hyperbolic fixed po<strong>in</strong>t p <strong>of</strong> saddle-type.<br />
2. The closure <strong>of</strong> <strong>the</strong> unstable manifold <strong>of</strong> p is an Hénon-like strange attractor A ′ .<br />
Let R α : x ↦→ x + α mod 1 a rotation over angle α ∈ (0, 1) \<br />
has a dense orbit <strong>in</strong> A = A ′ ¡ × 1 .<br />
. Then <strong>the</strong> product F = T × R<br />
Pro<strong>of</strong>: We claim that it is sufficient to prove <strong>the</strong> follow<strong>in</strong>g:<br />
(∗) Given two open sets U, V <strong>in</strong> A , <strong>the</strong>re exists k ∈ ¡ such that F k (U) ∩ V ≠ ∅.<br />
Indeed, given ε > 0 <strong>the</strong>re is a f<strong>in</strong>ite number <strong>of</strong> open sets V j , j ∈ J, <strong>of</strong> <strong>the</strong> form V j =<br />
V j ′ × (s j − ε, s j + ε) that cover A , where V j ′ ⊂ 2 is an open ball <strong>of</strong> radius ε. Let U 0 = U.<br />
Assum<strong>in</strong>g (∗), it follows that F k 1<br />
(U 0 ) <strong>in</strong>tersects V 1 for some k 1 ¡ ∈ . Def<strong>in</strong>e U 1 as <strong>the</strong> image<br />
under F −k 1<br />
<strong>of</strong> this <strong>in</strong>tersection. This process can be repeated for all j ∈ J. After this we<br />
restart <strong>the</strong> whole process with ε replaced by ε/2, ε/4, ε/8, . . ., ε/2 m , . . . , each time obta<strong>in</strong><strong>in</strong>g<br />
an open set U m such that U m ⊂ U m+1 . The <strong>in</strong>tersection ∩ m∈ U m gives an <strong>in</strong>itial po<strong>in</strong>t for a<br />
dense orbit as desired.<br />
Next, let us prove (∗). Without loss <strong>of</strong> generality, assume that U = U ′ × (r − δ, r + δ) <strong>and</strong><br />
V = V ′ × (s − ε, s + ε) for some δ, ε > 0, where U ′ , V ′ are open sets <strong>in</strong> A ′ . First, for fixed<br />
ε > 0 we note that given r, s ¡ ∈ 1 <strong>the</strong>re exists an <strong>in</strong>creas<strong>in</strong>g sequence {n 1 , n 2 , . . .} such that<br />
R n j<br />
α (r) ∈ (s − ε, s + ε), where 0 < n 1 < N <strong>and</strong> n j+1 − n j < N for all j, with N <strong>in</strong>dependent <strong>of</strong><br />
r <strong>and</strong> s. As W u (p) is dense <strong>in</strong> A ′ , <strong>the</strong>re exists a po<strong>in</strong>t q ′ ∈ W u (p) ∩ V ′ . Consider a preimage<br />
u = T −l (q ′ ) such that u <strong>and</strong> its first N iterates are close to p. By cont<strong>in</strong>uity, <strong>the</strong>re are<br />
open sets Z 0 , Z 1 , . . . , Z N<br />
around u, T (u), . . . , T N (u) whose images under T l , T l−1 , . . . , T l−N<br />
are conta<strong>in</strong>ed <strong>in</strong> V ′ .<br />
Now, <strong>the</strong>re exists a po<strong>in</strong>t x ∈ U ′ ∩ W u (p) belong<strong>in</strong>g to a dense orbit <strong>and</strong> also hav<strong>in</strong>g a<br />
positive Lyapunov exponent, such that T m (x) ∈ Z 0 for some m ∈ ¡ . It is no restriction to<br />
assume that, for some m ∈ ¡ , <strong>the</strong> image T m (U ′ ) <strong>in</strong>tersects all Z j , j = 0, 1, 2, . . . , N. Indeed,<br />
<strong>in</strong> <strong>the</strong> o<strong>the</strong>r case <strong>the</strong> Lyapunov exponent could not be positive.<br />
S<strong>in</strong>ce T l−j (Z j ∩ T m (U ′ )) ⊂ V ′ for j = 0, . . . , N, one has<br />
T l+m−j (U ′ ) ∩ V ′ ≠ ∅ (17)<br />
14
for all j = 0, . . . , N. To arrange that some <strong>of</strong> <strong>the</strong> iterates Rα<br />
l+m−j (r) lie <strong>in</strong>side <strong>the</strong> <strong>in</strong>terval<br />
(s − ε, s + ε), observe that l + m is <strong>in</strong> between two consecutive values n i <strong>and</strong> n i+1 for some<br />
n i as above. This implies that <strong>the</strong>re exists j with ≤ j ≤ N such that l + m − j = n i , which,<br />
toge<strong>the</strong>r with (17), yields that T l+m−j (U) ∩ V ≠ ∅.<br />
0.2<br />
0.15<br />
0.1<br />
0.05<br />
0<br />
0.2 0.25 0.3 0.35 0.4<br />
Figure 6: Diagram <strong>of</strong> <strong>the</strong> fully coupled system (7) <strong>in</strong> <strong>the</strong> (α, ε)-plane, for a = 1.25, b = 0.3, µ = 0.01<br />
<strong>and</strong> δ = 0.6/(2π). Accord<strong>in</strong>g to <strong>the</strong> values <strong>of</strong> <strong>the</strong> Lyapunov exponents, we <strong>in</strong>terpret as follows:<br />
doma<strong>in</strong>s <strong>of</strong> periodic attractors (code 1, yellow), <strong>of</strong> <strong>quasi</strong>-periodic attractors (code 2, blue), <strong>of</strong> Hénonlike<br />
attractors (code 3, red) <strong>and</strong> <strong>of</strong> ‘<strong>quasi</strong>-periodic Hénon-like’ attractors (codes 4, light blue, <strong>and</strong><br />
5, green). For more details see Section 2.3 <strong>and</strong> Section 4.<br />
2.3.2 Conjectural results<br />
Numerical experiments with <strong>the</strong> map (2) suggest that attractors like A persist for small (<strong>and</strong><br />
perhaps, not so small) values <strong>of</strong> (ε, δ). See figures 3 <strong>and</strong> 4. Figure 1 (C) gives evidence that<br />
<strong>quasi</strong>-periodic Hénon-like attractors occur <strong>in</strong> a relatively large part <strong>of</strong> <strong>the</strong> parameter plane<br />
(code 4, light blue). In this numerical context, <strong>quasi</strong>-periodic Hénon-like are <strong>in</strong>dicated by<br />
<strong>the</strong> fact that one positive, one negative, <strong>and</strong> one zero Lyapunov exponent is detected. We<br />
assume that <strong>the</strong> Lyapunov exponents are ordered as<br />
Λ 1 ≥ Λ 2 ≥ Λ 3 .<br />
A remarkable difference between <strong>the</strong> skew-product case (2) <strong>and</strong> <strong>the</strong> fully coupled case (7) is<br />
that a zero Lyapunov exponent practically never occurs, compare Figure 6 <strong>and</strong> see Section 4.2.<br />
Indeed, any small perturbation µ ≠ 0 has <strong>the</strong> effect <strong>of</strong> shift<strong>in</strong>g <strong>the</strong> value <strong>of</strong> Λ 2 away from<br />
zero. Its modulus rema<strong>in</strong>s small, but <strong>the</strong> sign may be ei<strong>the</strong>r positive or negative, where both<br />
cases occur. Plots <strong>of</strong> <strong>the</strong> attractors visually look <strong>the</strong> same <strong>in</strong> <strong>the</strong> three cases Λ 2 = 0, Λ 2 < 0<br />
<strong>and</strong> Λ 2 > 0, when µ is small <strong>and</strong> <strong>the</strong> o<strong>the</strong>r parameters are kept fixed. It rema<strong>in</strong>s to be<br />
clarified which dynamical <strong>and</strong> geometrical properties characterise <strong>the</strong> strange attractors <strong>in</strong><br />
each <strong>of</strong> <strong>the</strong> three cases. In any case, it seems that <strong>the</strong> way <strong>in</strong> which <strong>the</strong> <strong>in</strong>variant manifolds<br />
<strong>of</strong> an <strong>in</strong>variant circle are folded, plays an essential role.<br />
15
PSfrag replacements<br />
p<br />
U ′<br />
∂ s<br />
q<br />
c<br />
∂ u<br />
d<br />
Figure 7: Segments ∂ s <strong>and</strong> ∂ u <strong>of</strong> <strong>the</strong> stable <strong>and</strong> unstable manifold, respectively, <strong>of</strong> a saddle fixed<br />
po<strong>in</strong>t p bound a region U, see text for more explanation.<br />
Remark 4. 1. We used <strong>the</strong> family (7) for <strong>the</strong> figures, expect<strong>in</strong>g that it is sufficiently rich<br />
for our purposes.<br />
2. When compar<strong>in</strong>g <strong>the</strong> Figures 1 (C) <strong>and</strong> 6, <strong>the</strong> ma<strong>in</strong> difference is that <strong>in</strong> <strong>the</strong> second<br />
case <strong>the</strong>re is no significant occurrence <strong>of</strong> Λ 2 = 0. Still we expect that <strong>in</strong> all cases <strong>the</strong><br />
attractor is <strong>the</strong> closure Cl (W u (C )) <strong>of</strong> a <strong>quasi</strong>-periodic <strong>in</strong>variant circle C <strong>of</strong> saddle-type.<br />
3. It seems that <strong>in</strong> <strong>the</strong> skew case (2), <strong>the</strong> phenomenon <strong>of</strong> an attractor with Λ 1 = 0 <strong>and</strong><br />
Λ 2 < 0, which is not an <strong>in</strong>variant circle is somewhat related to ‘nonchaotic strange<br />
attractors’, compare with [28, 15, 17, 22, 29]. See also Section 4.3 for fur<strong>the</strong>r discussion<br />
on this topic.<br />
Interest<strong>in</strong>gly, t<strong>in</strong>y perturbations away from <strong>the</strong> skew case seem<strong>in</strong>gly give rise to a <strong>quasi</strong>periodic<br />
Hénon-like attractor, so with Λ 1 > 0.<br />
3 Pro<strong>of</strong>s<br />
3.1 Bas<strong>in</strong>s <strong>of</strong> attraction <strong>and</strong> <strong>quasi</strong>-periodic <strong>in</strong>variant circles<br />
In this section we give pro<strong>of</strong>s <strong>of</strong> Theorem 2 (next section) <strong>and</strong> Theorem 3 (Section 3.1.2).<br />
3.1.1 The Tangerman-Szewc argument generalised<br />
Let K : 2 → 2 be a dissipative diffeomorphism hav<strong>in</strong>g a saddle fixed po<strong>in</strong>t p = (x 0 , y 0 ).<br />
Suppose <strong>the</strong> stable <strong>and</strong> unstable manifolds W s (p) <strong>and</strong> W u (p) <strong>in</strong>tersect transversally at <strong>the</strong><br />
homocl<strong>in</strong>ic po<strong>in</strong>t q ∈ W s (p) ∩ W u (p), see Figure 7. Also assume that W u (p) is bounded as a<br />
subset <strong>of</strong><br />
2 . The Tangerman-Szewc Theorem (see e.g. [31, Appendix 3]) states that <strong>the</strong> bas<strong>in</strong><br />
<strong>of</strong> attraction <strong>of</strong> <strong>the</strong> closure <strong>of</strong> W u (p) conta<strong>in</strong>s <strong>the</strong> open region U ′ bounded by <strong>the</strong> two arcs<br />
∂ s ⊂ W s (p) <strong>and</strong> ∂ u ⊂ W u (p) with extremes p <strong>and</strong> q, see Figure 7. This argument is used to<br />
prove existence <strong>of</strong> strange attractors (<strong>in</strong> particular, with non-trivial bas<strong>in</strong> <strong>of</strong> attraction) near<br />
homocl<strong>in</strong>ic tangencies <strong>of</strong> a saddle fixed po<strong>in</strong>t <strong>of</strong> a dissipative diffeomorphism, cf. [26, 39, 42].<br />
We first prove Theorem 2 for ε = 0. This is a straightforward generalisation <strong>of</strong> <strong>the</strong><br />
above Tangerman-Szewc Theorem. For small ε, <strong>the</strong> result is obta<strong>in</strong>ed by us<strong>in</strong>g persistence <strong>of</strong><br />
normally hyperbolic <strong>in</strong>variant manifolds [20, Theorem 1.1] <strong>and</strong> two transversality lemmas.<br />
16
Pro<strong>of</strong> <strong>of</strong> Theorem 2. Consider <strong>the</strong> circle C α = C α,0 , <strong>in</strong>variant under map P α,0 <strong>in</strong> (6). The<br />
manifolds W u (C α ) <strong>and</strong> W s (C α ) are given by W u (p) × ¡ 1 <strong>and</strong> W s (p) × ¡ 1 , respectively. They<br />
<strong>in</strong>tersect transversally at a circle H = {q} × ¡ 1 , consist<strong>in</strong>g <strong>of</strong> po<strong>in</strong>ts homocl<strong>in</strong>ic to C α .<br />
Consider <strong>the</strong> two arcs ∂ s ⊂ W s (p) <strong>and</strong> ∂ u ⊂ W u (p) with extremes p <strong>and</strong> q (Figure 7).They<br />
bound an open set U ′ ⊂<br />
2 . Def<strong>in</strong>e D s <strong>and</strong> D u to be <strong>the</strong> portions <strong>of</strong> stable, <strong>and</strong> unstable<br />
manifold <strong>of</strong> C α , respectively, given by<br />
D s = ∂ s × ¡ 1 ⊂ W s (C α ) <strong>and</strong> D u = ∂ u × ¡ 1 ⊂ W u (C α ).<br />
Both surfaces D s <strong>and</strong> D u are compact, <strong>and</strong> <strong>the</strong>ir union forms <strong>the</strong> boundary <strong>of</strong> <strong>the</strong> open region<br />
U = U ′ × 1 , which ¡ is topologically a <strong>solid</strong> <strong>torus</strong>.<br />
The volume <strong>of</strong> U decreases under iteration <strong>of</strong> P α,0 . Denot<strong>in</strong>g by meas(·) <strong>the</strong> Lebesgue<br />
measure both on 2 <strong>and</strong> on 2 × 1 , due to condition 2 ¡ <strong>in</strong> Theorem 2 we have<br />
∫<br />
∫<br />
meas(Pα,0 n (U)) = 2π dxdy = 2π |det DK n | dxdy ≤ 2πκ n meas(U ′ ).<br />
K n (U ′ )<br />
U ′<br />
This implies that <strong>the</strong> forward evolution <strong>of</strong> every po<strong>in</strong>t (x, y, θ) ∈ U approaches <strong>the</strong> boundary<br />
<strong>of</strong> P n α,0 (U): dist ( P n α,0(x, y, θ), ∂P n α,0(U) ) → 0 as n → +∞.<br />
Indeed, suppose that this does not hold. Then <strong>the</strong>re exists a ϱ > 0 such that for all n <strong>the</strong>re<br />
exists N > n such that <strong>the</strong> ball with centre Pα,0(x, N y, θ) <strong>and</strong> radius ϱ > 0 is conta<strong>in</strong>ed <strong>in</strong>side<br />
Pα,0 N (U). But this would contradict <strong>the</strong> fact that meas(P α,0 n (U)) → 0 as n → +∞.<br />
The boundary <strong>of</strong> Pα,0 n (U) also consists <strong>of</strong> two portions <strong>of</strong> stable <strong>and</strong> unstable manifold <strong>of</strong><br />
C :<br />
∂Pα,0 n (U) = P α,0 n (Ds ) ∪ Pα,0 n (Du ).<br />
The diameter <strong>of</strong> P n α,0(D s ) tends to zero as n → +∞, because all po<strong>in</strong>ts <strong>in</strong> D s are attracted<br />
to <strong>the</strong> circle C α . S<strong>in</strong>ce W u (C α ) is bounded, all evolutions start<strong>in</strong>g <strong>in</strong> U are bounded <strong>and</strong><br />
approach W u (C α ), that is,<br />
dist(P n α,0 (x, y, θ), P n α,0 (Du )) → 0 as n → +∞<br />
for all (x, y, θ) ∈ U. This implies that ω(x, y, θ) ⊂ Cl (W u (C α )) for all (x, y, θ) ∈ U.<br />
To extend this result to small perturbations P α,ε <strong>of</strong> P α,0 , <strong>the</strong> follow<strong>in</strong>g transversality<br />
lemmas are used.<br />
Lemma 8. [33, Chap. 7] Consider a map f : V → M, where V <strong>and</strong> M are C r -<br />
differentiable manifolds <strong>and</strong> f is C r . Suppose V is compact, W ⊂ M is a closed C r -<br />
submanifold <strong>and</strong> f is transversal to W at V (notation: f ⋔ W ). Then f −1 (W ) is a C r -<br />
submanifold <strong>of</strong> codimension codim V (f −1 (W )) = codim M (W ). Fur<strong>the</strong>r suppose that <strong>the</strong>re is<br />
a neighbourhood <strong>of</strong> f(∂ V ) ∪ ∂ W disjo<strong>in</strong>t from f(V ) ∩ W , where ∂ V <strong>and</strong> ∂ W are <strong>the</strong> boundaries<br />
<strong>of</strong> V <strong>and</strong> W . Then any map g : V → M, sufficiently C r -close to f, is also transversal to W ,<br />
<strong>and</strong> <strong>the</strong> two submanifolds g −1 (W ) <strong>and</strong> f −1 (W ) are diffeomorphic.<br />
Lemma 9. [19, Section 3.2] Let V 1 , V 2 , <strong>and</strong> M be C r -differentiable manifolds <strong>and</strong> consider<br />
two <strong>diffeomorphisms</strong> f i : V i → M, i = 1, 2. Then f 1 ⋔ f 2 if <strong>and</strong> only if f 1 × f 2 ⋔ ∆,<br />
where f 1 × f 2 : V 1 × V 2 → M × M is <strong>the</strong> product map <strong>and</strong> ∆ ⊂ M × M is <strong>the</strong> diagonal:<br />
∆ = {(y, y) | y ∈ M}.<br />
Fix r ¡ ∈ <strong>and</strong> take ε < ε r , where ε r is given <strong>in</strong> Proposition 1. Then <strong>the</strong> map P α,ε has<br />
an r-normally hyperbolic <strong>in</strong>variant circle C α,ε <strong>of</strong> saddle type. Fur<strong>the</strong>rmore, <strong>the</strong> manifolds<br />
W u (C α,ε ), W s (C α,ε ), <strong>and</strong> C α,ε are C r -close to W u (C α ), W s (C α ), <strong>and</strong> C α . We now show<br />
17
that <strong>the</strong> two manifolds W u (C α,ε ), W s (C α,ε ) still <strong>in</strong>tersect transversally. To apply Lemma 8<br />
we restrict to two suitable compact subsets A u ⊂ W u (C α ) <strong>and</strong> A s ⊂ W s (C α ) as follows.<br />
Consider <strong>the</strong> segments pc ⊂ W u (p) <strong>and</strong> pd ⊂ W s (p) <strong>in</strong> Figure 7. Def<strong>in</strong>e<br />
A u = pc × ¡ 1 , A s = pd × ¡ 1 .<br />
In this way, <strong>the</strong> circle H is <strong>the</strong> <strong>in</strong>tersection <strong>of</strong> <strong>the</strong> manifolds A u <strong>and</strong> A s , bounded away from<br />
<strong>the</strong>ir boundaries. Consider <strong>the</strong> <strong>in</strong>clusions i : A u → M <strong>and</strong> j : A s → M. By <strong>the</strong> closeness <strong>of</strong><br />
W u (C α ) to W u (C α,ε ) <strong>the</strong>re exists a C r -diffeomorphism h : A u → A u ε ⊂ W u (C α,ε ) such that<br />
<strong>the</strong> map i is C r -close to i ε ◦ h, where i ε : A u ε → M is <strong>the</strong> <strong>in</strong>clusion [30, Section 2.6]. Similarly,<br />
<strong>the</strong>re exists a diffeomorphism k : A s → A s ε ⊂ W s (C α,ε ) such that <strong>the</strong> map j is C r -close j ε ◦ k,<br />
where j ε : A s ε → M is <strong>the</strong> <strong>in</strong>clusion. By Lemma 9 <strong>the</strong> map i × j : Au × A s → M × M is<br />
transversal to <strong>the</strong> diagonal ∆. For ε small, <strong>the</strong> map (i ε ◦ h) × (j ε ◦ k) : A u × A s → M × M is<br />
C r -close to i × j:<br />
A u × A s i×j<br />
−−−→ M × M<br />
⏐<br />
h×k↓<br />
A u ε × A s i ε×j ε<br />
ε −−−→ M × M.<br />
S<strong>in</strong>ce ∆ is closed <strong>and</strong> A u × A s is compact, Lemma 8 implies that <strong>the</strong>re exists an ε ∗ , with<br />
0 < ε ∗ < ε r , such that (i ε ◦ h) × (j ε ◦ k) ⋔ ∆ for ε < ε ∗ . Fur<strong>the</strong>rmore, <strong>the</strong> submanifolds<br />
(i × j) −1 (∆) <strong>and</strong><br />
(<br />
(iε ◦ h) × (j ε ◦ k) ) −1<br />
(∆)<br />
are diffeomorphic. We also have that ( (i ε ◦ h) × (j ε ◦ k) ) −1 (∆) is diffeomorphic to A u ε ∩ As ε ,<br />
<strong>and</strong> (i × j) −1 (∆) = A u ∩ A s = H .<br />
This shows that <strong>the</strong> <strong>in</strong>tersection H ε = A u ε ∩ As ε is diffeomorphic to H . Def<strong>in</strong>e Du ε as <strong>the</strong><br />
part <strong>of</strong> W u (C α,ε ) bounded by <strong>the</strong> <strong>in</strong>variant circle C α,ε <strong>and</strong> <strong>the</strong> circle <strong>of</strong> homocl<strong>in</strong>ic po<strong>in</strong>ts H ε .<br />
Def<strong>in</strong>e Dε s = k(D s ) similarly. Then <strong>the</strong> manifolds Dε u ⊂ W u (C α,ε ) <strong>and</strong> Dε s ⊂ W s (C α,ε ) form<br />
<strong>the</strong> boundary <strong>of</strong> an open region U ⊂ M homeomorphic to a <strong>torus</strong>. By <strong>the</strong> closeness <strong>of</strong> <strong>the</strong><br />
perturbed manifolds W s (C α,ε ) <strong>and</strong> W u (C α,ε ) to <strong>the</strong> unperturbed W s (C ) <strong>and</strong> W u (C ), both U<br />
<strong>and</strong> W u (C α,ε ) are bounded. Also notice that P α,ε is dissipative: by tak<strong>in</strong>g ε ∗ small enough,<br />
we ensure that |det(DF (x, y, θ))| < ˜c < 1 for all ε < ε ∗ <strong>and</strong> (x, y, θ) <strong>in</strong> U. Therefore, all<br />
forward evolutions beg<strong>in</strong>n<strong>in</strong>g at po<strong>in</strong>ts (x, y, θ) ∈ U rema<strong>in</strong> bounded. Like <strong>in</strong> <strong>the</strong> first part<br />
<strong>of</strong> <strong>the</strong> pro<strong>of</strong>, one has<br />
ω(x, y, θ) ⊂ Cl (W u (C α,ε ))<br />
for all (x, y, θ) ∈ U, α ∈ [0, 1] <strong>and</strong> ε < ε ∗ .<br />
3.1.2 An application <strong>of</strong> kam <strong>the</strong>ory<br />
So far, we did not discuss <strong>the</strong> dynamics <strong>in</strong> <strong>the</strong> saddle <strong>in</strong>variant circle C α,ε <strong>of</strong> map P α,ε<br />
<strong>in</strong> (6). Generically, <strong>the</strong> dynamics on C α,ε is <strong>of</strong> Morse-Smale type. In this case, <strong>the</strong> circle<br />
consists <strong>of</strong> <strong>the</strong> union <strong>of</strong> <strong>the</strong> unstable manifold <strong>of</strong> some periodic saddle. Theorem 3 describes<br />
a complementary case, for which <strong>the</strong> dynamics is <strong>quasi</strong>-periodic. Fix τ > 2 <strong>and</strong> def<strong>in</strong>e <strong>the</strong><br />
set <strong>of</strong> Diophant<strong>in</strong>e frequencies D γ by<br />
{<br />
D γ = α ∈ [0, 1] |<br />
∣ α − p }<br />
q ∣ ≥ γq−τ for all ¡ p, q ∈ , q ≠ 0 , (18)<br />
where γ > 0. S<strong>in</strong>ce we will apply a version <strong>of</strong> <strong>the</strong> KAM Theorem hold<strong>in</strong>g for non-conservative,<br />
f<strong>in</strong>itely differentiable systems (see [5, Chap. 5] <strong>and</strong> [6]), a certa<strong>in</strong> amount <strong>of</strong> smoothness <strong>of</strong><br />
<strong>the</strong> circle C α,ε is needed, depend<strong>in</strong>g on <strong>the</strong> Diophant<strong>in</strong>e condition specified <strong>in</strong> (18). Therefore<br />
we require that <strong>the</strong> perturbed family <strong>of</strong> maps P α,ε is C n , for n large enough.<br />
18
Pro<strong>of</strong> <strong>of</strong> Theorem 3. Consider map P α,0 <strong>in</strong> (6), <strong>and</strong> let p = (x 0 , y 0 ) be a saddle fixed po<strong>in</strong>t<br />
<strong>of</strong> <strong>the</strong> diffeomorphism K. The <strong>in</strong>variant circle C α,0 = {p} × ¡ 1 <strong>of</strong> P α,0 can be trivially seen<br />
as a graph over ¡ 1 :<br />
C α,0 = { (x 0 , y 0 , θ) | θ ∈ ¡<br />
1 } .<br />
Fix r ¡ ∈ large enough <strong>and</strong> ε < ε r , where ε r is taken as <strong>in</strong> Proposition 1. By <strong>the</strong> C r -closeness<br />
<strong>of</strong> C α,0 <strong>and</strong> C α,ε (Proposition 1), <strong>the</strong> circle C α,ε <strong>of</strong> P α,ε can be written as a C r -graph ¡ over 1 :<br />
C α,ε = { (x ε (θ), y ε (θ), θ) ∈ 2 × ¡ 1 | θ ∈ ¡<br />
1 } , (19)<br />
where x ε ¡ : 1 → , x ε (θ) = x 0 + O(ε), <strong>and</strong> similarly for y ε (θ). So <strong>the</strong> restriction <strong>of</strong> P α,ε to<br />
C α,ε has <strong>the</strong> follow<strong>in</strong>g form<br />
P α,ε<br />
∣<br />
∣Cα,ε<br />
: C α,ε → C α,ε , P α,ε (θ) = θ + α + εg ε (x 0 , y 0 , θ, α) + O(ε 2 ).<br />
By (19), we may consider P α,ε as a map on ¡ 1 . Fix γ > 0, τ > 3 <strong>and</strong> take D γ as <strong>in</strong> (18). For<br />
α ∈ D γ , <strong>the</strong> map P α,ε can be averaged repeatedly over <strong>the</strong> circle, putt<strong>in</strong>g <strong>the</strong> θ-dependency<br />
<strong>in</strong>to terms <strong>of</strong> higher order <strong>in</strong> ε, compare [8, Proposition 2.7] <strong>and</strong> [11, Section 4]. After such<br />
changes <strong>of</strong> variables, P α,ε is brought <strong>in</strong>to <strong>the</strong> normal form<br />
P α,ε (θ) = θ + α + c(α, ε) + O(ε r+1 ).<br />
In fact, it is convenient to consider α as a variable, <strong>and</strong> to def<strong>in</strong>e <strong>the</strong> cyl<strong>in</strong>der maps<br />
P ε : ¡ 1 × [0, 1] → ¡ 1 × [0, 1], P ε (θ, α) = (P α,ε (θ), α)<br />
R : ¡ 1 × [0, 1] → ¡ 1 × [0, 1],<br />
R(θ, α) = (R α (θ), α),<br />
where R α : ¡ 1 → ¡ 1 is <strong>the</strong> rigid rotation <strong>of</strong> an angle α. We now apply a version <strong>of</strong> <strong>the</strong> KAM<br />
Theorem, hold<strong>in</strong>g for non-conservative, f<strong>in</strong>itely differentiable systems (see e.g. [5, Chap. 5]<br />
<strong>and</strong> [6]), to <strong>the</strong> family <strong>of</strong> <strong>diffeomorphisms</strong> P ε . There exists an <strong>in</strong>teger m with 1 ≤ m < r <strong>and</strong><br />
a C m -map<br />
Φ ε : ¡ 1 × [0, 1] → ¡ 1 × [0, 1], Φ ε (θ, α) = (θ + εA(θ, α, ε), α + εB(α, ε)), (20)<br />
such that <strong>the</strong> restriction <strong>of</strong> Φ ε to ¡ 1 × D γ makes <strong>the</strong> follow<strong>in</strong>g diagram commute:<br />
1 × ¡<br />
R<br />
D γ −−−→ 1 × D ¡ γ<br />
↑ ↑<br />
Φ ⏐⏐<br />
ε Φ ⏐⏐<br />
ε<br />
1 × D ¡<br />
P ε<br />
γ −−−→ 1 × D ¡ γ .<br />
The differentiability <strong>of</strong> Φ ε restricted to ¡ 1 × D γ is <strong>of</strong> Whitney type. S<strong>in</strong>ce P α,ε<br />
∣<br />
∣Cα,ε<br />
is C m -<br />
conjugate to a rigid rotation on ¡ 1 , <strong>the</strong> circle C α,ε is <strong>in</strong> fact C m . This proves parts 1 <strong>and</strong> 2<br />
<strong>of</strong> <strong>the</strong> Theorem.<br />
Fur<strong>the</strong>rmore, <strong>the</strong> constant γ <strong>in</strong> (18) can be taken equal to ε r . This gives that <strong>the</strong> measure<br />
<strong>of</strong> <strong>the</strong> complement <strong>of</strong> D γ <strong>in</strong> [0, 1] is <strong>of</strong> order ε r as ε → 0.<br />
3.2 Hénon-like attractors do exist<br />
Our pro<strong>of</strong> <strong>of</strong> Theorem 4 is based on a result <strong>of</strong> Díaz-Rocha-Viana [14]. We beg<strong>in</strong> by stat<strong>in</strong>g<br />
this result.<br />
19
3.2.1 Perturbations <strong>of</strong> multimodal families<br />
Two def<strong>in</strong>itions from [14, Section 5.2] are <strong>in</strong>troduced now. For more <strong>in</strong>formation about <strong>the</strong><br />
term<strong>in</strong>ology, we refer to [24, Sections II.5, II.6].<br />
Def<strong>in</strong>ition 4. Let J ⊂ be a compact <strong>in</strong>terval. Fix d ≥ 1, k ≥ 3, a ∗ ∈ , <strong>and</strong> an <strong>in</strong>terval<br />
<strong>of</strong> parameter values U = [a − , a + ], with a ∗ ∈ Int U. A C k -family <strong>of</strong> maps M a : J → J, with<br />
a ∈ U, is called a d-family if it satisfies <strong>the</strong> follow<strong>in</strong>g conditions:<br />
1. Invariance: M a ∗(J) ⊂ Int(J);<br />
2. Nondegenerate critical po<strong>in</strong>ts: M a ∗ has d critical po<strong>in</strong>ts {c 1 , . . . , c d } def<br />
= Cr M a ∗ that satisfy<br />
M ′′<br />
a ∗(c i) ≠ 0 for all i <strong>and</strong> M a ∗(c i ) ≠ c j for all i, j;<br />
3. Negative Schwarzian derivative: SM a ∗ < 0 for all x ≠ c i , where<br />
Sf(x) = f ′′′ (x)<br />
f ′ (x) − 3 ( ) f ′′ 2<br />
(x)<br />
;<br />
2 f ′ (x)<br />
4. Topological mix<strong>in</strong>g: There exists an <strong>in</strong>terval J ′ ⊂ Int(J) such that M a ∗(J) = M a ∗(J ′ ) = J ′<br />
<strong>and</strong> such that for any open <strong>in</strong>tervals J 1 , J 2 <strong>in</strong> J ′ <strong>the</strong>re exists n 0 such that<br />
M n a ∗(J 1) ∩ J 2 ≠ ∅ for all n ≥ n 0 ;<br />
5. Pre<strong>periodicity</strong>: for each 1 ≤ i ≤ d <strong>the</strong>re exists m i such that p i = M m i<br />
a ∗ (c i) is a (repell<strong>in</strong>g)<br />
periodic po<strong>in</strong>t <strong>of</strong> M a ∗;<br />
6. Genericity <strong>of</strong> unfold<strong>in</strong>g: For all c i ∈ Cr M a ∗, denote by c i (a) <strong>and</strong> p i (a) <strong>the</strong> cont<strong>in</strong>uations<br />
<strong>of</strong> c i <strong>and</strong> p i , respectively, for a close to a ∗ . Then<br />
d (<br />
M<br />
m i<br />
a<br />
da<br />
(c i(a)) − p i (a) ) ≠ 0 at a = a ∗ .<br />
Next we <strong>in</strong>troduce <strong>the</strong> notion <strong>of</strong> η-perturbation <strong>of</strong> a d-family M a , with a ∈ U <strong>and</strong> d ≥ 1 fixed.<br />
Def<strong>in</strong>ition 5. Fix σ > 0 <strong>and</strong> consider <strong>the</strong> family M a obta<strong>in</strong>ed by extend<strong>in</strong>g M a as follows:<br />
M a : J × I σ → J × I σ ,<br />
M a (x, y) def<br />
=(M a (x), 0). (21)<br />
Also denote by M <strong>the</strong> map<br />
M : U × J × I σ → J × I σ ,<br />
M(a, x, y) def<br />
= M a (x, y) = (M a (x), 0).<br />
Given a C k -family <strong>of</strong> <strong>diffeomorphisms</strong><br />
for a k ≥ 3, denote by G its extension<br />
G a : J × I σ → J × I σ , a ∈ J,<br />
G : U × J × I σ → J × I σ ,<br />
G(a, x, y) def<br />
= G a (x, y).<br />
Then G is called a η-perturbation <strong>of</strong> <strong>the</strong> d-family {M a } a if<br />
‖M − G‖ C k ≤ η,<br />
where ‖·‖ C k denotes <strong>the</strong> C k -norm over U × J × I σ .<br />
20
The follow<strong>in</strong>g proposition is used <strong>in</strong> <strong>the</strong> sequel to prove existence <strong>of</strong> Hénon-like attractors for<br />
<strong>the</strong> map (2). See [2, 3, 26, 32, 37, 39, 42] for similar results.<br />
Proposition 10. [14, Theorem 5.2] Let {M a } a be a d-family <strong>and</strong> p a periodic po<strong>in</strong>t <strong>of</strong><br />
M a ∗. Then <strong>the</strong>re exist η > 0, ā <strong>and</strong> χ > 0 such that, given any η-perturbation {G a } a <strong>of</strong><br />
{M a } a <strong>the</strong> follow<strong>in</strong>g holds.<br />
1. For all a with |a − a ∗ | < ā <strong>the</strong> map G a has a periodic po<strong>in</strong>t p a which is <strong>the</strong> cont<strong>in</strong>uation<br />
<strong>of</strong> <strong>the</strong> periodic po<strong>in</strong>t (p, 0) <strong>of</strong> <strong>the</strong> map M a <strong>in</strong> (21).<br />
2. There exists a set S, conta<strong>in</strong>ed <strong>in</strong> <strong>the</strong> <strong>in</strong>terval [a ∗ − ā, a ∗ + ā] ⊂ U, with meas(S) > χ,<br />
such that for all a ∈ S <strong>the</strong>re exists z ∈ W u (p a ) satisfy<strong>in</strong>g:<br />
(a) <strong>the</strong> orbit {G n a(z) | n ≥ 0} is dense <strong>in</strong> Cl (W u (Orb(p a )));<br />
(b) G a has a positive Lyapunov exponent at z, i.e., <strong>the</strong>re exist k > 0, λ > 1 <strong>and</strong> v ≠ 0<br />
such that ‖DG n a(z)v‖ ≥ kλ n for all n ≥ 0;<br />
(c) <strong>the</strong>re exist w ≠ 0 such that ‖DG n a(z)w‖ → 0 as n → ±∞.<br />
3.2.2 Multimodal families aris<strong>in</strong>g from powers <strong>of</strong> <strong>the</strong> Logistic map<br />
The pro<strong>of</strong> <strong>of</strong> Theorem 4, which we present <strong>in</strong> this section, is based on three facts. First,<br />
suppose that a ∗ ∈ [0, 2] is such that <strong>the</strong> quadratic family Q a (x) = 1 − ax 2 <strong>in</strong> (13) is a d-<br />
def<br />
family <strong>in</strong> <strong>the</strong> sense <strong>of</strong> Def<strong>in</strong>ition 4, with d = 1. Then for all n ≥ 1 <strong>the</strong> family M a = Q n a<br />
given by <strong>the</strong> n-th iterate <strong>of</strong> Q a is a d-family for some d ≤ 2 n . Second, for all η 1 > 0, <strong>the</strong><br />
composition <strong>of</strong> an η 1 -perturbation <strong>of</strong> Q a with an η 1 -perturbation <strong>of</strong> Q n a is an η 2-perturbation<br />
<strong>of</strong> Q n+1<br />
a , where η 2 = C(n)η 1 <strong>and</strong> C(n) is a positive constant depend<strong>in</strong>g on n. Third, for each<br />
n > q ≥ 1 <strong>and</strong> for each (α, δ) ∈ A q/n , <strong>the</strong> asymptotic dynamics <strong>of</strong> T α,δ,a,ε is described by a<br />
map that turns out to be an η-perturbation <strong>of</strong> <strong>the</strong> d-family M a , with η = O(ε). Moreover,<br />
M a has a periodic po<strong>in</strong>t p such that its analytic cont<strong>in</strong>uation <strong>in</strong> <strong>the</strong> family T α,δ,a,ε possesses<br />
a transversal homocl<strong>in</strong>ic <strong>in</strong>tersection. Application <strong>of</strong> Proposition 10 to <strong>the</strong> po<strong>in</strong>t p concludes<br />
<strong>the</strong> pro<strong>of</strong>.<br />
In <strong>the</strong> next lemma we show that M a is a d-family. For each ã ∈ [0, 2) <strong>the</strong>re exists a β > 0<br />
such that for all a with a ∈ [0, ã] <strong>the</strong> <strong>in</strong>terval J = [−1−β, 1+β] ⊂ satisfies Q a (J) ⊂ Int(J).<br />
In <strong>the</strong> sequel, it is always assumed that <strong>the</strong> family Q a is def<strong>in</strong>ed on such an <strong>in</strong>terval J, <strong>and</strong><br />
that <strong>the</strong> values <strong>of</strong> a we consider are such that Q a (J) ⊂ Int(J).<br />
Lemma 11. Suppose a ∗ ∈ [0, 2) def<br />
= U is such that <strong>the</strong> quadratic family<br />
Q a : J → J, Q a (x) = 1 − ax 2<br />
satisfies hypo<strong>the</strong>ses 4 <strong>and</strong> 5 <strong>of</strong> Def<strong>in</strong>ition 4. Then for all n ≥ 1 <strong>the</strong>re exists d ≥ 1 such that<br />
<strong>the</strong> family<br />
def<br />
M a : J → J, M a = Q n a<br />
is a d-family with d ≤ 2 n − 1 critical po<strong>in</strong>ts.<br />
Pro<strong>of</strong>. Take a ∗ as above. We first prove <strong>the</strong> case n = 1, that is, Q a : J a → J a is a 1-family.<br />
Conditions 1, 2, 3 <strong>of</strong> Def<strong>in</strong>ition 4 are obviously satisfied by Q a . Condition 6 will now be<br />
proved. By conditions 4 <strong>and</strong> 5 (assumed by hypo<strong>the</strong>sis), Q a ∗ is a Misiurewicz map [25], i.e.,<br />
it has no periodic attractor <strong>and</strong> c ∉ ω(c), where c = 0 is <strong>the</strong> critical po<strong>in</strong>t <strong>of</strong> Q a ∗. Moreover,<br />
21
y [24, Theorem III.6.3] <strong>the</strong> map Q a ∗ is Collet-Eckmann (see e.g. [24, Section V.4]), that is,<br />
<strong>the</strong>re exist constants κ > 0 <strong>and</strong> λ > 1 such that<br />
∣ d<br />
∣∣∣<br />
∣dx a ∗(Q a ∗(c)) ≥ κλ n for all n ≥ 0. (22)<br />
Therefore, by comb<strong>in</strong><strong>in</strong>g [38, Theorem 3] with <strong>the</strong> Collet-Eckmann condition (22) we get<br />
lim<br />
n→∞<br />
d<br />
da Qn a (c) | a=a ∗<br />
d<br />
dx Qn−1 a ∗ (Q a<br />
∗(c))<br />
> 0. (23)<br />
Assume Q k a ∗(c) = p, with p periodic (<strong>and</strong> repell<strong>in</strong>g) under Q a∗. By p(a) denote <strong>the</strong> cont<strong>in</strong>uation<br />
<strong>of</strong> p for a close to a ∗ . Then, for all n sufficiently large,<br />
d<br />
da Qn a (c) | a=a ∗ = ∂Qn−k a<br />
∂a<br />
= ∂<br />
∂a Qn−k a<br />
= d da<br />
(Qk a ∗(c)) | a=a ∗ +∂Qn−k a<br />
∂x (Qk a ∗(c)) | d<br />
a=a ∗<br />
(p) | a=a ∗ + ∂<br />
d<br />
(p) | a=a ∗<br />
(<br />
Q<br />
n−k<br />
a (p(a)) ) + ∂<br />
∂x Qn−k a<br />
∂x Qn−k a ∗<br />
da Qk a (c) | a=a ∗=<br />
[<br />
p(a) + Q<br />
k<br />
da<br />
a (c) − p(a) ] | a=a ∗=<br />
(p) d [<br />
Q<br />
k<br />
da a (c) − p(a) ] | a=a ∗ .<br />
(24)<br />
The po<strong>in</strong>t Qa<br />
n−k (p(a)) belongs to a hyperbolic periodic orbit, that varies smoothly with <strong>the</strong><br />
parameter a. Therefore, its derivative with respect to a (which is <strong>the</strong> first term <strong>in</strong> <strong>the</strong> last<br />
equality) is uniformly bounded <strong>in</strong> n. On <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>,<br />
d<br />
dx Qn−1 a ∗<br />
(Q a ∗(c)) = ∂<br />
∂x Qn−k a ∗<br />
(p) d<br />
dx Qk−1 a ∗<br />
(Q a ∗(c)).<br />
Therefore, by (22), (23), <strong>and</strong> (24) we conclude that<br />
d<br />
da<br />
0 < lim<br />
Qn a (c) | [<br />
d<br />
a=a ∗<br />
n→∞ d<br />
dx Qn−1 a (Q ∗ a ∗(c)) = da Q<br />
k<br />
a (c) − p(a) ] a=a ∗<br />
d<br />
dx Qk−1 a (Q ∗ a ∗(c)) . (25)<br />
This proves that Q a satisfies condition 6 <strong>of</strong> Def<strong>in</strong>ition 4.<br />
We now show that <strong>the</strong> n-th iterate M a <strong>of</strong> <strong>the</strong> quadratic map is a d-family for all n > 1<br />
<strong>and</strong> for some d ≤ 2 n . For simplicity, we denote Q a ∗ by Q for <strong>the</strong> rest <strong>of</strong> this pro<strong>of</strong>. Condition<br />
1 holds for M a ∗ s<strong>in</strong>ce it holds for Q a ∗. Condition 3 follows from <strong>the</strong> fact that <strong>the</strong> composition<br />
<strong>of</strong> maps with negative Schwarzian derivative also has negative Schwarzian derivative, see<br />
e.g. [24, II.6]. Condition 4 is obviously satisfied.<br />
Condition 2 is now proved by <strong>in</strong>duction on n, where <strong>the</strong> case n = 1 is obvious. Obviously,<br />
<strong>the</strong> set Cr M a ∗ <strong>of</strong> critical po<strong>in</strong>ts <strong>of</strong> M a ∗ has card<strong>in</strong>ality d ≤ 2 n − 1. Moreover,<br />
Cr M a ∗ = Q ( −1 Cr Q n−1) n−1<br />
⋃<br />
∪ Cr Q = (Q −j )(Cr Q). (26)<br />
Suppose that condition 2 holds for a given n ≥ 1. We first show that<br />
j=0<br />
(Q n+1 ) ′′ (x) ≠ 0 for all x ∈ Cr Q n+1 . (27)<br />
By (26), if x ∈ Cr Q n+1 <strong>the</strong>n ei<strong>the</strong>r x = c, or Q(x) ∈ Cr Q n . If x = c <strong>the</strong>n<br />
(Q n+1 ) ′′ (x) = (Q n ) ′ (Q(c)) · (Q) ′′ (c). (28)<br />
22
The second factor is nonzero. If <strong>the</strong> first factor is zero, <strong>the</strong>n<br />
0 = (Q n ) ′ (Q(c)) = Q ′ (Q n (c)) . . . Q ′ (Q(c)).<br />
Therefore <strong>the</strong>re exists j such that Q j (c) = c, so that c is an attract<strong>in</strong>g periodic po<strong>in</strong>t <strong>of</strong><br />
Q. But this contradicts <strong>the</strong> fact that Q is Misiurewicz, so that (28) is nonzero. The o<strong>the</strong>r<br />
possibility is that c ≠ x <strong>and</strong> Q(x) ∈ Cr Q n . In this case,<br />
(Q n+1 ) ′′ (x) = (Q n ) ′′ (Q(x)) · Q ′ (x) 2 ,<br />
which is nonzero. Indeed, Q ′ (x) ≠ 0, o<strong>the</strong>rwise x = c. Moreover (Q n ) ′′ (Q(x)) ≠ 0 by <strong>the</strong><br />
<strong>in</strong>duction hypo<strong>the</strong>ses s<strong>in</strong>ce <strong>the</strong> critical po<strong>in</strong>ts <strong>of</strong> Q n are nondegenerate. This proves (27),<br />
from which <strong>the</strong> first part <strong>of</strong> condition 2 follows.<br />
We now prove, aga<strong>in</strong> argu<strong>in</strong>g by contradiction, that<br />
Q n+1 (x) ≠ y for all x, y ∈ Cr Q n+1 .<br />
Suppose that <strong>the</strong>re exist x, y ∈ Cr Q n+1 such that Q n+1 (x) = y. By (26) <strong>the</strong>re exist i <strong>and</strong> j<br />
such that Q i (x) = Q j (y) = c, where 0 ≤ i, j ≤ n. This would imply that<br />
Q n+1+j−i (c) = Q j (Q n+1 (x)) = Q j (y) = c,<br />
with n + 1 + j − i ≥ 1 <strong>and</strong>, <strong>the</strong>refore, c would be an attract<strong>in</strong>g periodic po<strong>in</strong>t <strong>of</strong> Q, which is<br />
impossible s<strong>in</strong>ce Q is Misiurewicz. Condition 2 is proved.<br />
To prove condition 5, fix y ∈ Cr M a ∗ <strong>and</strong> j ≥ 0 such that Q j (y) = c. S<strong>in</strong>ce c is preperiodic<br />
for Q by hypo<strong>the</strong>sis, <strong>the</strong>re exists k ≥ 1 such that Q j+k (y) = p, where p is periodic under Q<br />
with period u ≥ 1. The orbit <strong>of</strong> y under M a ∗ is, except for a f<strong>in</strong>ite number <strong>of</strong> <strong>in</strong>itial iterates,<br />
a subset <strong>of</strong> <strong>the</strong> orbit <strong>of</strong> p under Q. This shows that y is preperiodic for M a ∗.<br />
To prove condition 6, take y ∈ Cr M a ∗, j, u, k <strong>and</strong> p ∈ J as <strong>in</strong> <strong>the</strong> pro<strong>of</strong> <strong>of</strong> condition 5.<br />
Then <strong>the</strong>re exist <strong>in</strong>tegers l <strong>and</strong> m, with 0 ≤ l < u <strong>and</strong> m ≥ 1, such that<br />
M m a ∗(y) = Qk+l (c) = Q l (p) ∈ Orb Q (p). (29)<br />
By condition 5 (assumed by hypo<strong>the</strong>sis) <strong>and</strong> by (29), <strong>the</strong> po<strong>in</strong>t z = Q l (p) is periodic (<strong>and</strong><br />
repell<strong>in</strong>g) under M a ∗. Denote by y(a), z(a), <strong>and</strong> p(a) <strong>the</strong> cont<strong>in</strong>uations <strong>of</strong> y, z, <strong>and</strong> p,<br />
respectively, for a close to a ∗ . In particular,<br />
We have to show that<br />
By <strong>the</strong> cha<strong>in</strong> rule we get<br />
d<br />
da Ql+k a<br />
Q j a (y(a)) = c <strong>and</strong> Ql a (p(a)) = z(a).<br />
d<br />
da [M m a (y(a)) − z(a)]| a=a ∗ ≠ 0. (30)<br />
(c) ∣ a=a ∗<br />
= ∂Ql a<br />
∂a (Qk a (c))∣ ∣<br />
a=a ∗<br />
+ ∂Ql a<br />
∂x (Qk a (c))∣ ∣<br />
dQ k a<br />
a=a ∗<br />
da (c)| a=a<br />
= ∗<br />
= ∂Ql a ∗<br />
∂a (p) + ∂Ql a ∗<br />
∂x (p) dQk a ∗<br />
da (c),<br />
d<br />
da Ql a(p(a)) ∣ a=a ∗<br />
= ∂Ql a ∗<br />
∂a (p) + ∂Ql a ∗<br />
∂x (p) d da p(a∗ ),<br />
where p = p(a ∗ ) = Q k a∗(c). Therefore,<br />
d<br />
da [M a m (y(a)) − z(a)]| a=a ∗ = d [<br />
Q<br />
k+l<br />
a (c) − Q l<br />
da<br />
a(p(a)) ]∣ ∣<br />
a=a ∗<br />
=<br />
= ∂Ql a ∗<br />
∂x (p) d [<br />
Q<br />
k<br />
da a (c) − p(a) ]∣ ∣<br />
a=a ∗<br />
.<br />
[<br />
The factor d da Q<br />
k<br />
a (c(a)) − p(a) ]∣ ∣<br />
a=a ∗<br />
is nonzero by (25). The same holds for <strong>the</strong> o<strong>the</strong>r factor,<br />
o<strong>the</strong>rwise p would be an attract<strong>in</strong>g periodic po<strong>in</strong>t <strong>of</strong> Q a ∗. This proves <strong>in</strong>equality (30).<br />
23
Proposition 10 does not provide a nontrivial bas<strong>in</strong> <strong>of</strong> attraction for <strong>the</strong> closure Cl (W u (p a )).<br />
We now show that, under <strong>the</strong> hypo<strong>the</strong>ses <strong>of</strong> Theorem 4, <strong>the</strong>re exists a periodic po<strong>in</strong>t p a for<br />
which a nontrivial bas<strong>in</strong> <strong>of</strong> attraction <strong>of</strong> Cl (W u (p a )) can be constructed. Therefore, <strong>in</strong> this<br />
case Cl (W u (p a )) is a Hénon-like attractor.<br />
Proposition 12. Consider <strong>the</strong> map {Ma ∗ } a = Q n a ∗, where Q a∗ satisfies <strong>the</strong> hypo<strong>the</strong>ses <strong>of</strong><br />
Lemma 11. There exist a periodic po<strong>in</strong>t p <strong>of</strong> Q a ∗, <strong>and</strong> positive constants η, ā <strong>and</strong> χ such that<br />
for any η-perturbation {G a } a <strong>of</strong> {M a } a = Q n a <strong>the</strong> follow<strong>in</strong>g holds.<br />
∗<br />
1. For all a with |a − a ∗ | < ā <strong>the</strong> map G a has a periodic po<strong>in</strong>t p a which is <strong>the</strong> cont<strong>in</strong>uation<br />
<strong>of</strong> <strong>the</strong> periodic po<strong>in</strong>t (p, 0) <strong>of</strong> <strong>the</strong> map M a <strong>in</strong> (21). Moreover, p a has a transversal<br />
homocl<strong>in</strong>ic <strong>in</strong>tersection.<br />
2. There exists a set S, conta<strong>in</strong>ed <strong>in</strong> <strong>the</strong> <strong>in</strong>terval [a ∗ − ā, a ∗ + ā] ⊂ U, with meas(S) > χ,<br />
such that for all a ∈ S <strong>the</strong> set Cl (W u (Orb(p a ))) is a Hénon-like attractor <strong>of</strong> <strong>the</strong> map<br />
G a .<br />
Pro<strong>of</strong>. To construct a non-trivial bas<strong>in</strong> <strong>of</strong> attraction for Cl (W u (Orb(p a ))), it is sufficient<br />
to f<strong>in</strong>d a periodic po<strong>in</strong>t p a <strong>of</strong> {G a } a that has a transversal homocl<strong>in</strong>ic <strong>in</strong>tersection. Then<br />
<strong>the</strong> bas<strong>in</strong> is provided by <strong>the</strong> Tangerman-Szewc Theorem (see Theorem 2 <strong>and</strong> subsequent<br />
remark). Indeed, for all η sufficiently small, all η-perturbations <strong>of</strong> <strong>the</strong> map Q ∗ a are uniformly<br />
dissipative. Moreover, <strong>the</strong> unstable manifold <strong>of</strong> p a is bounded, s<strong>in</strong>ce it is bounded for Q a ∗<br />
<strong>and</strong> s<strong>in</strong>ce <strong>the</strong> <strong>in</strong>variant manifolds <strong>of</strong> a map depend cont<strong>in</strong>uously on <strong>the</strong> map [26, Prop. 7.1].<br />
Therefore, <strong>the</strong> second part <strong>of</strong> <strong>the</strong> proposition follows from <strong>the</strong> first part, toge<strong>the</strong>r with <strong>the</strong><br />
Tangerman-Szewc argument <strong>and</strong> Proposition 10.<br />
To prove <strong>the</strong> first part, we claim that <strong>the</strong> map Q a ∗ has a periodic po<strong>in</strong>t p belong<strong>in</strong>g to<br />
a non-degenerate homocl<strong>in</strong>ic orbit. Indeed, if <strong>the</strong> claim is true, <strong>the</strong>n for η sufficiently small<br />
<strong>and</strong> for a close to a ∗ , any η-perturbation <strong>of</strong> M a possesses a periodic po<strong>in</strong>t p a which is <strong>the</strong><br />
analytic cont<strong>in</strong>uation <strong>of</strong> p <strong>and</strong> such that p a has a transverse homocl<strong>in</strong>ic <strong>in</strong>tersection. The<br />
latter property aga<strong>in</strong> follows from cont<strong>in</strong>uous dependence <strong>of</strong> <strong>the</strong> <strong>in</strong>variant manifolds on <strong>the</strong><br />
map [26, Prop. 7.1].<br />
To prove <strong>the</strong> claim that Q a ∗ has a periodic po<strong>in</strong>t p belong<strong>in</strong>g to a non-degenerate homocl<strong>in</strong>ic<br />
orbit we first show that <strong>the</strong>re exists a po<strong>in</strong>t y 0 belong<strong>in</strong>g to a degenerate homocl<strong>in</strong>ic<br />
orbit <strong>of</strong> Q a ∗. S<strong>in</strong>ce <strong>the</strong> critical po<strong>in</strong>t c <strong>of</strong> Q a ∗ is preperiodic, <strong>the</strong>re exist positive <strong>in</strong>tegers k, h<br />
such that Q k a ∗(c) = y 0 <strong>and</strong> y 0 is periodic with period h. The unstable manifold <strong>of</strong> any periodic<br />
po<strong>in</strong>t <strong>of</strong> Q a ∗ is <strong>the</strong> whole core [1 − a ∗ , 1], s<strong>in</strong>ce Q a ∗ is topologically mix<strong>in</strong>g. Therefore, s<strong>in</strong>ce<br />
<strong>the</strong> critical po<strong>in</strong>t c belongs to W u (y 0 ), by tak<strong>in</strong>g preimages <strong>of</strong> c, a po<strong>in</strong>t q can be found such<br />
that q ∈ Wloc u (y 0), Q l a∗(q) = c <strong>and</strong> Ql+k a (q) = y ∗ 0 for some <strong>in</strong>teger l > 0. This means that y 0<br />
belongs to a degenerate homocl<strong>in</strong>ic orbit <strong>of</strong> Q a ∗.<br />
We now prove that <strong>the</strong>re exists a periodic po<strong>in</strong>t p <strong>of</strong> Q a ∗ hav<strong>in</strong>g a non-degenerate homocl<strong>in</strong>ic<br />
orbit. This is achieved by exam<strong>in</strong><strong>in</strong>g a power <strong>of</strong> Q a ∗ for which all po<strong>in</strong>ts <strong>of</strong> <strong>the</strong> orbit<br />
<strong>of</strong> y 0 are fixed. Denote by Orb Qa ∗(y 0 ) = {y j | j = 0, . . . , h − 1} <strong>the</strong> orbit <strong>of</strong> y 0 , under Q a ∗,<br />
where y j = Q j a ∗(y 0). Let m be <strong>the</strong> smallest multiple <strong>of</strong> h which is larger than k, <strong>and</strong> denote<br />
f def<br />
= Q m a ∗. Then, f(c) belongs to Orb Q a ∗(y 0 ) <strong>and</strong> all po<strong>in</strong>ts y j are fixed for f. We can assume<br />
that<br />
f ′ (y j ) > 1 for all y j ∈ Orb Qa ∗(y 0 ) (31)<br />
by tak<strong>in</strong>g f 2 <strong>in</strong>stead <strong>of</strong> f if necessary.<br />
Brouwer’s fixed po<strong>in</strong>t Theorem <strong>and</strong> cont<strong>in</strong>uity arguments ensure <strong>the</strong> existence <strong>of</strong> a fixed<br />
po<strong>in</strong>t p <strong>of</strong> f, a critical po<strong>in</strong>t c ′ <strong>of</strong> f, <strong>and</strong> an <strong>in</strong>terval I = (c ′ − δ, c ′ ) such that:<br />
1. f ′ (p) < −1;<br />
24
2. c ′ lies <strong>in</strong> <strong>the</strong> <strong>in</strong>terval (y, p);<br />
3. f is monotonically <strong>in</strong>creas<strong>in</strong>g <strong>in</strong> I;<br />
4. p falls <strong>in</strong> <strong>the</strong> <strong>in</strong>terval f(I).<br />
The configuration <strong>of</strong> c, c ′ <strong>and</strong> y = f(c) with<strong>in</strong> <strong>the</strong> graph <strong>of</strong> f looks like <strong>the</strong> sketch <strong>in</strong> Figure 8,<br />
<strong>in</strong> <strong>the</strong> case y < c <strong>and</strong> f ′′ (c) > 0 (<strong>the</strong> o<strong>the</strong>r comb<strong>in</strong>ations <strong>of</strong> <strong>the</strong> sign <strong>of</strong> y − c <strong>and</strong> f ′′ (c) are<br />
treated similarly). S<strong>in</strong>ce f is topologically mix<strong>in</strong>g, <strong>the</strong> <strong>in</strong>terval f(I) is conta<strong>in</strong>ed <strong>in</strong> <strong>the</strong><br />
PSfrag replacements<br />
p<br />
f(c)<br />
y<br />
Figure 8: Graph <strong>of</strong> <strong>the</strong> map f from <strong>the</strong> pro<strong>of</strong> <strong>of</strong> Proposition 12. Only <strong>the</strong> relevant branches <strong>of</strong><br />
<strong>the</strong> graph are plotted, <strong>in</strong> relation to <strong>the</strong> fixed po<strong>in</strong>ts y, p <strong>and</strong> <strong>the</strong> critical po<strong>in</strong>ts c <strong>and</strong> c ′ . See <strong>the</strong><br />
text for details.<br />
unstable manifold <strong>of</strong> p. Therefore p belongs to a homocl<strong>in</strong>ic orbit O.<br />
Moreover, <strong>the</strong> homocl<strong>in</strong>ic orbit O is non-degenerate. Indeed, if this was not <strong>the</strong> case, <strong>the</strong>n<br />
<strong>the</strong>re would exist a critical po<strong>in</strong>t c ′′ <strong>of</strong> f belong<strong>in</strong>g to O, so that f j (c ′′ ) = p for some j ¡ ∈ .<br />
However, accord<strong>in</strong>g to (26), <strong>and</strong> s<strong>in</strong>ce c is preperiodic, <strong>the</strong> orbit <strong>of</strong> c ′′ under Q a ∗ eventually<br />
l<strong>and</strong>s <strong>in</strong>side Orb Qa ∗(y 0 ). It follows that p ∈ Orb Qa ∗(y 0 ), which is absurd, s<strong>in</strong>ce f ′ (p) < −1<br />
whereas (31) holds.<br />
In <strong>the</strong> next lemma we show that <strong>the</strong> composition <strong>of</strong> a small perturbation <strong>of</strong> <strong>the</strong> map<br />
Q a (x, y) = (Q a (x), 0) (we use here <strong>the</strong> notation <strong>of</strong> Def<strong>in</strong>ition 5) with a small perturbation <strong>of</strong><br />
Q n a (x, y) = (Qn a (x), 0) yields a small perturbation <strong>of</strong> Qn+1 a (x, y). As <strong>in</strong> Def<strong>in</strong>ition 5, denote by<br />
Q, Q n : [0, 2]×J×I → J×I <strong>the</strong> functions Q(a, x, y) = (Q a (x), 0) <strong>and</strong> Q n (a, x, y) = (Q n a(x), 0),<br />
respectively.<br />
Lemma 13. For each η > 0 <strong>the</strong>re exists a ζ > 0 such that for all F, G : [0, 2] × J × I → J × I<br />
such that<br />
‖G − Q‖ C 3 < ζ <strong>and</strong> ‖F − Q n ‖ C 3 < ζ, (32)<br />
c ′<br />
p<br />
c<br />
we have ∥ ∥G ◦ F − Q<br />
n+1 ∥ ∥<br />
C 3<br />
< η. (33)<br />
Pro<strong>of</strong>. Write<br />
G(a, x, y) =<br />
( )<br />
Qa (x) + g 1 (a, x, y)<br />
g 2 (a, x, y)<br />
<strong>and</strong> F (a, x, y) =<br />
( )<br />
Q<br />
n<br />
a (x) + f 1 (a, x, y)<br />
.<br />
f 2 (a, x, y)<br />
25
Then<br />
G ◦ F (a, x, y) −<br />
( ) Q<br />
n+1<br />
a (x)<br />
=<br />
0<br />
( −2a(f1 (a, x, y)) 2 − 2af 1 (a, x, y)Q n a (x) + g 1(<br />
a, ˜f1 (a, x, y), f 2 (a, x, y) ) )<br />
(<br />
g 2 a, ˜f1 (a, x, y), f 2 (a, x, y) ) ,<br />
where ˜f 1 (a, x, y) = Q n a(x) + f 1 (a, x, y). The C 3 -norm <strong>of</strong> <strong>the</strong> terms −2a(f 1 (a, x, y)) 2 <strong>and</strong><br />
−2af 1 (a, x, y)Q n a (x) is bounded by a constant times <strong>the</strong> C3 -norm <strong>of</strong> f 1 . We now estimate <strong>the</strong><br />
norm <strong>of</strong> ˜g 1 , def<strong>in</strong>ed by<br />
˜g 1 (x 0 , x 1 , x 2 ) = g 1 (a, ˜f 1 (a, x, y), f 2 (a, x, y)).<br />
Denote x 0 = a, x 1 = x, <strong>and</strong> x 2 = y. Then any second order derivative <strong>of</strong> ˜g 1 is a sum <strong>of</strong> terms<br />
<strong>of</strong> <strong>the</strong> follow<strong>in</strong>g type:<br />
∂ 2 g 1 ∂ ˜f k ∂g 1 ∂ 2 ˜fk<br />
,<br />
,<br />
∂x j x k ∂x l ∂x k ∂x j x l<br />
where we put ˜f 2 = f 2 to simplify <strong>the</strong> notation. For <strong>the</strong> third order derivatives a similar<br />
property holds. S<strong>in</strong>ce <strong>the</strong> C 3 -norm <strong>of</strong> ˜f k is bounded, we get that each term <strong>in</strong> <strong>the</strong> third order<br />
derivative <strong>of</strong> ˜g 1 is bounded by a constant times <strong>the</strong> C 3 -norm <strong>of</strong> <strong>the</strong> g j . This concludes <strong>the</strong><br />
pro<strong>of</strong>.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 4. The Theorem will be first proved for a ∗ < 2. The case a ∗ = 2<br />
follows by choos<strong>in</strong>g ano<strong>the</strong>r value ā ∗ < 2 sufficiently close to 2. Fix a ∗ ∈ [0, 2) verify<strong>in</strong>g <strong>the</strong><br />
hypo<strong>the</strong>ses <strong>of</strong> Lemma 11. To beg<strong>in</strong> with, we consider <strong>the</strong> case (α, δ) ∈ Int A 1 , <strong>the</strong> <strong>in</strong>terior <strong>of</strong><br />
<strong>the</strong> tongue <strong>of</strong> period one. Then <strong>the</strong> Arnol ′ d family A α,δ on 1 has two hyperbolic fixed po<strong>in</strong>ts<br />
¡<br />
θ1 s (attract<strong>in</strong>g) <strong>and</strong> θ1 r (repell<strong>in</strong>g), see [13, Section 1.14]. The θ-coord<strong>in</strong>ate <strong>of</strong> both po<strong>in</strong>ts<br />
depends on <strong>the</strong> choice <strong>of</strong> (α, δ) ∈ Int A 1 . So for all ¡ θ ∈ 1 with θ ≠ θ1 r , <strong>the</strong> orbit <strong>of</strong> θ under<br />
A α,δ converges to θ1 s . This means that <strong>the</strong> manifold<br />
Θ 1 = { }<br />
(x, y, θ) ∈ 2 × 1 | θ = θ1<br />
s ¡ ⊂ 2 1<br />
¡ ×<br />
is <strong>in</strong>variant <strong>and</strong> attract<strong>in</strong>g under T α,δ,a,ε . Denote by G a,1 <strong>the</strong> restriction <strong>of</strong> T α,δ,a,ε to Θ 1 :<br />
G a,1 : Θ 1 → Θ 1 , (x, y, θ s 1 ) ↦→ (1 − ax2 + εf 1 , εg 1 , θ s 1 ),<br />
where f 1 = f(a, x, y, θ s 1 , α, δ) <strong>and</strong> similarly for g 1. S<strong>in</strong>ce Q a ∗(J) ⊂ Int(J), <strong>the</strong>re exists a<br />
constant σ > 0 such that for all ε sufficiently small <strong>and</strong> all a close enough to a ∗ ,<br />
G a,1 (J × I σ × {θ s 1 }) ⊂ Int(J × I σ × {θ s 1 })<br />
T α,δ,a,ε<br />
(<br />
J × Iσ × (¡ 1 \ {θ r 1 })) ⊂ Int ( J × I σ × (¡ 1 \ {θ r 1 })) . (34)<br />
S<strong>in</strong>ce Θ 1 is diffeomorphic to 2 , we consider G a,1 as a map <strong>of</strong> 2 . Then G a,1 , is an<br />
η-perturbation <strong>of</strong> <strong>the</strong> quadratic family Q a (x), where η = O(ε). We now apply Proposition<br />
12 to <strong>the</strong> family G a,1 . Let p 0 be <strong>the</strong> periodic po<strong>in</strong>t <strong>of</strong> M a ∗ as given by Proposition 12.<br />
For all ε sufficiently small <strong>the</strong>re exists a constant ā > 0 <strong>and</strong> a set S <strong>of</strong> positive Lebesgue<br />
measure, conta<strong>in</strong>ed <strong>in</strong> <strong>the</strong> <strong>in</strong>terval [a ∗ − ā, a ∗ + ā], such that <strong>the</strong> follow<strong>in</strong>g holds. For all<br />
a ∈ [a ∗ − ā, a ∗ + ā], G a,1 has a saddle periodic po<strong>in</strong>t ¯p which is <strong>the</strong> cont<strong>in</strong>uation <strong>of</strong> <strong>the</strong> po<strong>in</strong>t<br />
p 0 . Fur<strong>the</strong>rmore, for all a ∈ S <strong>the</strong> closure à = Cl ( W u (Orb Ga,1 (¯p)) ) is a Hénon-like attractor<br />
<strong>of</strong> G a,1 conta<strong>in</strong>ed <strong>in</strong>side Θ 1 . The po<strong>in</strong>t p = (¯p, θ1 s) is a saddle periodic po<strong>in</strong>t <strong>of</strong> <strong>the</strong> map T α,δ,a,ε,<br />
<strong>and</strong> W u (Orb Tα,δ,a,ε (p)) = W u (Orb Ga,1 (¯p)) × {θ1 s}. Therefore A = Cl (W u (p)) = Ã × {θs 1 } is a<br />
26<br />
<strong>and</strong>
Hénon-like attractor <strong>of</strong> T α,δ,a,ε . Moreover, <strong>the</strong> bas<strong>in</strong> <strong>of</strong> attraction <strong>of</strong> Cl (W u (p)) has nonempty<br />
<strong>in</strong>terior <strong>in</strong> 2 ¡ × 1 because <strong>of</strong> (34). This proves <strong>the</strong> claim for (α, δ) ∈ Int A 1 .<br />
We pass to <strong>the</strong> case <strong>of</strong> higher period tongues. Suppose that (α, δ) ∈ Int A q/n , with<br />
n > q ≥ 1. Then A α,δ has (at least) two hyperbolic periodic orbits<br />
Orb(θ s 1 ) = {θs 1 , θs 2 , . . . , θs n }<br />
Orb(θ r 1 ) = {θr 1 , θr 2 , . . . , θr n }<br />
attract<strong>in</strong>g, <strong>and</strong><br />
repell<strong>in</strong>g.<br />
For j = 1, . . . , n, denote by Θ j <strong>the</strong> manifold<br />
Θ j = { (x, y, θ) ∈ 2 × ¡ 1 ) | θ = θ s j}<br />
,<br />
<strong>and</strong> def<strong>in</strong>e maps G j as <strong>the</strong> restriction <strong>of</strong> T α,δ,a,ε to Θ j :<br />
G j : Θ j → Θ j+1 for j = 1, . . . , n − 1<br />
G n : Θ n → Θ 1 ,<br />
where<br />
(x, y, θ1 s ) G j<br />
↦→ (Q a (x) + εf j , εg j , θj+1 s ), for j = 1, . . . , n − 1<br />
(x, y, θn) s ↦→ Gn<br />
(Q a (x) + εf n , εg n , θ1).<br />
s<br />
Here, f j = f(a, x, y, θj s, α, δ). The manifold Θ 1 is <strong>in</strong>variant <strong>and</strong> attract<strong>in</strong>g under <strong>the</strong> n-th<br />
iterate <strong>of</strong> <strong>the</strong> map T α,δ,a,ε . For all (x, y, θ) <strong>in</strong> <strong>the</strong> complement <strong>of</strong> <strong>the</strong> set<br />
<strong>the</strong> asymptotic dynamics is given by <strong>the</strong> map<br />
{(x, y, θ) | θ ∈ Orb(θ r 1)} ,<br />
def<br />
G a,1,...n = G n ◦ G n−1 ◦ · · · ◦ G 1 .<br />
Notice that each <strong>of</strong> <strong>the</strong> G j ’s is an η j -perturbation <strong>of</strong> <strong>the</strong> family Q a <strong>in</strong> <strong>the</strong> sense <strong>of</strong> Def<strong>in</strong>ition 5,<br />
where η j = Bε <strong>and</strong> B can be chosen uniform on θj s (<strong>and</strong>, <strong>the</strong>refore, on (α, δ)).<br />
Let p 0 be <strong>the</strong> periodic po<strong>in</strong>t <strong>of</strong> M a ∗ as given by Proposition 12. Then (p 0 , 0) is a saddle<br />
periodic po<strong>in</strong>t for <strong>the</strong> map M a def<strong>in</strong>ed as <strong>in</strong> (21). Take η, ā, <strong>and</strong> χ as <strong>in</strong> Proposition 12. By<br />
<strong>in</strong>ductive application <strong>of</strong> Lemma 13 <strong>the</strong>re exists an ¯ε > 0 depend<strong>in</strong>g on η <strong>and</strong> n such that<br />
‖G a,1,...n − Q n ‖ C 3 < η,<br />
for all (α, δ) ∈ Int A q/n <strong>and</strong> all |ε| < ¯ε. That is, G a,1,...n is an η-perturbation <strong>of</strong> M a for all q<br />
with 1 ≤ q < n <strong>and</strong> all (α, δ, a, ε) with<br />
(α, δ) ∈ A q/n ,<br />
ε ∈ [−¯ε, ¯ε].<br />
By Proposition 12 <strong>the</strong>re exist an ā > 0 <strong>and</strong> a set S conta<strong>in</strong>ed <strong>in</strong> <strong>the</strong> <strong>in</strong>terval [a ∗ − ā, a ∗ + ā]<br />
such that meas(S) ≥ χ <strong>and</strong> <strong>the</strong> follow<strong>in</strong>g holds. For all a ∈ [a ∗ −ā, a ∗ +ā] <strong>the</strong> map G a,1,...,n has<br />
a periodic po<strong>in</strong>t ¯p a which is <strong>the</strong> cont<strong>in</strong>uation <strong>of</strong> <strong>the</strong> periodic po<strong>in</strong>t (p 0 , 0) <strong>of</strong> M a . Moreover,<br />
for all a ∈ S <strong>the</strong> closure à = Cl ( W u (Orb Ga,1,...,n (¯p a )) ) is a Hénon-like attractor <strong>of</strong> G a,1,...,n ,<br />
conta<strong>in</strong>ed <strong>in</strong>side Θ 1 .<br />
To f<strong>in</strong>ish <strong>the</strong> pro<strong>of</strong>, observe that p a = (¯p a , θ1 s) is a saddle periodic po<strong>in</strong>t <strong>of</strong> T α,δ,a,ε. The<br />
set A = Cl ( W u (Orb Tα,δ,a,ε (p a )) ) is compact <strong>and</strong> <strong>in</strong>variant under T α,δ,a,ε . To show that A<br />
has a dense orbit, fix parameter values as provided by Proposition 12 applied to G a,1,...n . Let<br />
z ∈ Θ 1 a po<strong>in</strong>t hav<strong>in</strong>g a dense orbit <strong>in</strong> Ã <strong>and</strong> satisfy<strong>in</strong>g properties (a)–(c) <strong>of</strong> Proposition 10.<br />
Then given η > 0 <strong>and</strong> a po<strong>in</strong>t<br />
q = T j α,δ,a,ε (q′ ) ∈ T j α,δ,a,ε (Ã × {θs 1 }), with 1 ≤ j ≤ n − 1,<br />
27
<strong>the</strong>re exists m > 0 such that dist(G m a,1,...n (z), q′ ) < η. By cont<strong>in</strong>uity <strong>of</strong> T j α,δ,a,ε<br />
, for all ϱ > 0<br />
<strong>the</strong>re exists η > 0 such that<br />
dist(T j α,δ,a,ε (q′′ ), T j α,δ,a,ε (q′ )) < ϱ for all q ′′ with dist(q ′′ , q ′ ) < η.<br />
We conclude that for all ϱ > 0 <strong>the</strong>re exists m > 0 such that<br />
dist(T j α,δ,a,ε (Gm a,1,...n (z)), T j α,δ,a,ε (q′ )) = dist(T j+mn<br />
α,δ,a,ε<br />
(z), q) < ϱ.<br />
This proves that <strong>the</strong> orbit <strong>of</strong> z under T α,δ,a,ε is dense <strong>in</strong> A . Properties (11) <strong>and</strong> (12) will now<br />
be proved. S<strong>in</strong>ce G a,1,...,n = T n α,δ,a,ε on Θ 1, for any m ∈ ¡ <strong>and</strong> any z ∈ A we have<br />
DT m α,δ,a,ε(z) = DT r α,δ,a,ε(G s a,1,...,n(z))DG s a,1,...,n(z),<br />
where s = m mod n <strong>and</strong> r = m − s. Take z as above <strong>and</strong> a vector v = (v x , v y , 0) ∈ T z A<br />
such that ∥ ∥ DG<br />
s<br />
a,1,...,n (z)v ∥ ∥ ≥ κλ s for all s, where κ > 0 <strong>and</strong> λ > 1 are constants. S<strong>in</strong>ce T r α,δ,a,ε<br />
is a diffeomorphism for all r = 1, . . . , s − 1 <strong>and</strong> G s a,1,...,n(z) belongs to <strong>the</strong> compact set A for<br />
all s ∈ ¡ , <strong>the</strong>n <strong>the</strong>re exists a constant c > 0 such that<br />
∥ DT<br />
m<br />
α,δ,a,ε (z)v ∥ ∥ =<br />
∥ ∥DT r<br />
α,δ,a,ε (G s a,1,...,n (z))DGs a,1,...,n (z)v∥ ∥ ≥ c<br />
∥ ∥DG s<br />
a,1,...,n (z)v ∥ ∥ ,<br />
where c is uniform <strong>in</strong> r. This proves property (11). Property (12) is proved similarly. This<br />
shows that <strong>the</strong> closure Cl (W u (p a )) is a Hénon-like attractor <strong>of</strong> T α,δ,a,ε .<br />
Remark 5. At <strong>the</strong> boundary <strong>of</strong> a tongue A q/n <strong>the</strong> Arnol ′ d family A α,δ has a saddle-node<br />
periodic po<strong>in</strong>t θ 1 . However, <strong>the</strong> bas<strong>in</strong> <strong>of</strong> attraction <strong>of</strong> Orb θ 1 still has nonempty <strong>in</strong>terior, so<br />
that <strong>the</strong> above conclusions hold for all (α, δ) <strong>in</strong> <strong>the</strong> closure Cl ( A q/n) .<br />
4 Numerical methods, results <strong>and</strong> <strong>in</strong>terpretation<br />
4.1 Methods <strong>and</strong> selection <strong>of</strong> parameters<br />
An important tool <strong>in</strong> <strong>the</strong> numerical exploration <strong>of</strong> dynamical systems consists <strong>of</strong> <strong>the</strong> computation<br />
<strong>of</strong> Lyapunov exponents. Let us take a three-dimensional map T as before, with an<br />
orbit {x j , j = 0, 1, 2, 3, . . .} . Follow<strong>in</strong>g [35] we start with three <strong>in</strong>dependent tangent vectors,<br />
<strong>of</strong> which <strong>the</strong> successive iterates under <strong>the</strong> derivative DT are computed, after some transient.<br />
At each step (or after a given number <strong>of</strong> steps to speed up <strong>the</strong> process) <strong>the</strong> vectors are<br />
orthonormalised.<br />
It is useful to <strong>in</strong>troduce Lyapunov sums, as we show now, for simplicity just consider<strong>in</strong>g <strong>the</strong><br />
iterates <strong>of</strong> one tangent vector. Let v 0 be <strong>the</strong> <strong>in</strong>itial vector <strong>and</strong> write v n = DT n x 0<br />
(v 0 ), i.e., <strong>the</strong><br />
tangent vector obta<strong>in</strong>ed after n iterations. Let ˆv n+1 = DT (x n )v n . Then v n+1 = ˆv n+1 /f n+1 ,<br />
where f n+1 = ‖ˆv n+1 ‖. The Lyapunov sum <strong>the</strong>n is def<strong>in</strong>ed as<br />
LS n =<br />
n∑<br />
log(f j ). (35)<br />
j=1<br />
The Lyapunov exponent <strong>the</strong>n is <strong>the</strong> average slope <strong>of</strong> <strong>the</strong> Lyapunov sum as n → ∞, that is,<br />
<strong>the</strong> average <strong>of</strong> <strong>the</strong> logarithmic rates <strong>of</strong> <strong>in</strong>crease <strong>of</strong> <strong>the</strong> length log(f j ).<br />
In <strong>the</strong> numerical procedure estimates are produced <strong>of</strong> <strong>the</strong> average slope <strong>of</strong> LS n up to<br />
some n for different values <strong>of</strong> n up to a maximal number <strong>of</strong> N iterates. The computations are<br />
stopped before N iterates <strong>in</strong> case <strong>of</strong> escape, or if a periodic orbit is detected, or if different<br />
28
estimates <strong>of</strong> <strong>the</strong> average co<strong>in</strong>cide with<strong>in</strong> a prescribed tolerance ρ. Typical values for N <strong>and</strong><br />
ρ <strong>in</strong> <strong>the</strong> present computations are 10 7 <strong>and</strong> 10 −6 , respectively.<br />
One <strong>of</strong> <strong>the</strong> major problems is to detect values <strong>of</strong> <strong>the</strong> Lyapunov exponents very close to<br />
zero. To this end several procedures have been proposed for obta<strong>in</strong><strong>in</strong>g <strong>the</strong> limit. Tak<strong>in</strong>g <strong>in</strong>to<br />
account that <strong>in</strong> <strong>the</strong> skew case <strong>the</strong> driv<strong>in</strong>g behaviour is <strong>quasi</strong>-periodic or periodic, a method <strong>of</strong><br />
successive filter<strong>in</strong>g <strong>and</strong> fitt<strong>in</strong>g, similar to [7] can be suitable. Ano<strong>the</strong>r method like MEGNO<br />
(see [12] for an exposition <strong>and</strong> examples <strong>and</strong> [23] for a problem which requires a massive use<br />
<strong>of</strong> it) is based <strong>of</strong> weighted averages <strong>and</strong> is very useful to detect small values <strong>of</strong> <strong>the</strong> Lyapunov<br />
exponent. However, presently we simply use <strong>the</strong> Lyapunov sum (35) because this will help<br />
to underst<strong>and</strong> <strong>the</strong> behaviour <strong>of</strong> <strong>the</strong> system <strong>in</strong> some elementary cases, see Section 4.3.<br />
To scan <strong>the</strong> behaviour <strong>of</strong> family T given by (7), several parameters have been fixed. We<br />
chose δ = 0.6 such that resonant zones <strong>of</strong> <strong>the</strong> Arnol ′ d family are not too narrow, while still<br />
most <strong>of</strong> <strong>the</strong> values <strong>of</strong> α give rise to <strong>quasi</strong>-periodic dynamics. Concern<strong>in</strong>g <strong>the</strong> parameters a<br />
<strong>and</strong> b <strong>of</strong> <strong>the</strong> Hénon family, we fixed b = 0.3 for historical reasons. It is <strong>the</strong> value used by<br />
Hénon [18], <strong>and</strong> it is a good compromise between dissipation <strong>and</strong> visibility <strong>of</strong> <strong>the</strong> folds <strong>of</strong><br />
<strong>the</strong> unstable manifold. It was also used <strong>in</strong> [34], where <strong>the</strong> various attractors for this value<br />
<strong>of</strong> b, for several values <strong>of</strong> a was studied, as well as <strong>the</strong> role <strong>of</strong> homocl<strong>in</strong>ic <strong>and</strong> heterocl<strong>in</strong>ic<br />
tangencies (later on <strong>in</strong> <strong>the</strong> literature known as ‘crises’). The value a = 1.25 corresponds to a<br />
periodic attractor <strong>of</strong> period 7 <strong>and</strong> allows for moderate values <strong>of</strong> ε <strong>in</strong> <strong>the</strong> forc<strong>in</strong>g before escape<br />
occurs. F<strong>in</strong>ally we selected <strong>the</strong> values µ = 0 <strong>and</strong> µ = 0.01 for <strong>the</strong> skew <strong>and</strong> <strong>the</strong> fully coupled<br />
case, respectively. O<strong>the</strong>r values <strong>of</strong> µ have also been <strong>in</strong>vestigated, see Section 4.2.<br />
Let Λ 1 ≥ Λ 2 ≥ Λ 3 be <strong>the</strong> three Lyapunov exponents. S<strong>in</strong>ce <strong>the</strong> family (7) is dissipative,<br />
<strong>the</strong> role <strong>of</strong> Λ 3 is not very relevant. It can only help to decide, <strong>in</strong> case <strong>of</strong> periodic or <strong>quasi</strong>periodic<br />
attractors, whe<strong>the</strong>r <strong>the</strong> normal behaviour is <strong>of</strong> nodal type (Λ 2 > Λ 3 ) or <strong>of</strong> focal type<br />
(Λ 2 = Λ 3 ). The major role is played by Λ 1 <strong>and</strong> Λ 2 <strong>and</strong> <strong>the</strong>ir position with respect to zero.<br />
In some cases it is also <strong>in</strong>terest<strong>in</strong>g to use a complementary tool to help to decide whe<strong>the</strong>r<br />
<strong>the</strong> attractor is <strong>quasi</strong>-periodic or has some ‘strange’ character. This occurs for small values<br />
<strong>of</strong> ε. In <strong>the</strong> skew case µ = 0 one may expect to have a period 7 <strong>in</strong>variant curve if (α, δ) is <strong>in</strong><br />
<strong>the</strong> <strong>quasi</strong>-periodic doma<strong>in</strong> <strong>and</strong> ε is sufficiently small. Similar behaviour can be expected for<br />
µ > 0 small, for a large relative measure set <strong>in</strong> (α, δ).<br />
The follow<strong>in</strong>g method has been used. Consider iterates <strong>of</strong> T 7 , after some transient, <strong>and</strong><br />
sort <strong>the</strong>m by <strong>the</strong> values <strong>of</strong> θ. If <strong>the</strong> attractor is an <strong>in</strong>variant curve (x(θ), y(θ)), <strong>the</strong>n <strong>the</strong><br />
variation <strong>of</strong> <strong>the</strong> components (x, y) can be estimated from <strong>the</strong> iterates. This variation has to<br />
rema<strong>in</strong> bounded when <strong>the</strong> number <strong>of</strong> iterates <strong>in</strong>creases <strong>and</strong> tend to <strong>the</strong> true variation (but see<br />
Section 4.3 for some wrong <strong>in</strong>terpretations <strong>of</strong> <strong>the</strong> results <strong>of</strong> <strong>the</strong>se computations). To recognise<br />
Hénon-like attractors for <strong>the</strong> fully coupled case µ > 0 a similar device is used. The values <strong>of</strong><br />
<strong>the</strong> angular component θ <strong>of</strong> <strong>the</strong> iterates, should cluster around <strong>the</strong> periodic orbit obta<strong>in</strong>ed <strong>in</strong><br />
<strong>the</strong> case µ = 0.<br />
4.2 Numerical results<br />
The diagrams <strong>in</strong> Figures 1 (C) <strong>and</strong> 6 are based on <strong>the</strong> values <strong>of</strong> <strong>the</strong> first two Lyapunov<br />
exponents Λ 1 <strong>and</strong> Λ 2 . To be more precise, <strong>in</strong> Figure 1 (C) code 1 (yellow) corresponds to<br />
0 > Λ 1 , code 2 (blue) to 0 = Λ 1 > Λ 2 , code 3 (red) to Λ 1 > 0 > Λ 2 <strong>and</strong> code 4 (light blue)<br />
to Λ 1 > 0 = Λ 2 . Typically we considered a Lyapunov exponent as equal to zero whenever<br />
|Λ j | < 10 −5 , j = 1, 2.<br />
In Figure 6 a new case appears, namely where Λ 1 > Λ 2 > 0. Here <strong>the</strong> value <strong>of</strong> Λ 2 is<br />
small but def<strong>in</strong>itely positive. It occurs for some regions <strong>of</strong> <strong>the</strong> (α, ε) parameter plane which,<br />
<strong>in</strong> <strong>the</strong> skew case µ = 0 <strong>of</strong> Figure 1 (C) seem<strong>in</strong>gly show <strong>quasi</strong>-periodic Hénon-like attractors.<br />
29
For <strong>the</strong>se parameter values we ma<strong>in</strong>ta<strong>in</strong> <strong>the</strong> code 4 (light blue). Fur<strong>the</strong>rmore, <strong>in</strong>side <strong>the</strong><br />
case Λ 1 > 0 > Λ 2 one has to dist<strong>in</strong>guish two subcases: one which corresponds to Hénon-like<br />
attractors (identified by <strong>the</strong> cluster<strong>in</strong>g <strong>of</strong> θ, as described before) <strong>and</strong> for which we keep <strong>the</strong><br />
code 3 (red), <strong>and</strong> ano<strong>the</strong>r with Λ 2 close to zero but def<strong>in</strong>itely negative. The latter can be<br />
seen as a perturbation <strong>of</strong> <strong>the</strong> <strong>quasi</strong>-periodic Hénon attractors. We use for <strong>the</strong>se <strong>the</strong> code 5<br />
(green). The existence <strong>of</strong> parameter values for which <strong>the</strong> Lyapunov exponents are positive<br />
has been recently also found <strong>in</strong> a quite different context, related to what can be considered<br />
as a discrete version <strong>of</strong> Lorenz attractor, see [16].<br />
Figure 9: Attractor <strong>of</strong> T as <strong>in</strong> (7) for (α, ε, µ) = (0.31, 0.13, 0.01), with two positive Lyapunov<br />
exponents: Λ 1 ≈ 0.29530 <strong>and</strong> Λ 2 ≈ 0.00016 (which is close to zero). We note that no visual<br />
difference is observed with attractors hav<strong>in</strong>g Λ 2 negative close to zero (fully coupled case µ > 0) or<br />
Λ 2 = 0 (skew product case µ = 0). The representation uses variables (u, v, w) similar to Figure 3.<br />
Interest<strong>in</strong>gly, no visual differences can be observed between <strong>the</strong>se attractors <strong>in</strong> <strong>the</strong> cases<br />
where Λ 1 > 0 <strong>and</strong> Λ 2 = 0 (<strong>in</strong> <strong>the</strong> skew case µ = 0) or where Λ 1 > 0 <strong>and</strong> Λ 2 is close to<br />
zero, <strong>and</strong> ei<strong>the</strong>r positive or negative (<strong>in</strong> <strong>the</strong> fully coupled model µ > 0). Figure 9 displays<br />
<strong>the</strong> detected attractor for (α, ε, µ) = (0.31, 0.13, 0.01) (<strong>in</strong> <strong>the</strong> region <strong>of</strong> code 4 <strong>in</strong> Figure 6).<br />
The plot uses variables (u, v, w) similar to Figure 3. Mov<strong>in</strong>g <strong>the</strong> parameters to (α, ε, µ) =<br />
(0.28, 0.13, 0.01) (code 5 region <strong>in</strong> Figure 6) or to (α, ε, µ) = (0.28, 0.13, 0.00) (code 4 region<br />
<strong>in</strong> Figure 1), <strong>in</strong> all <strong>the</strong>se cases <strong>the</strong> attractor looks quite similar. Fur<strong>the</strong>r study is needed to<br />
clarify <strong>the</strong> geometric differences, by consider<strong>in</strong>g <strong>the</strong> expected saddle-type <strong>in</strong>variant circle <strong>and</strong><br />
its <strong>in</strong>variant manifolds.<br />
Compar<strong>in</strong>g Figures 1 (C) <strong>and</strong> 6 we observe:<br />
1) The region code 1 (periodic attractors, yellow) <strong>in</strong> Figure 1 (C) is essentially preserved<br />
<strong>in</strong> Figure 6, where more periodic attractors were detected near <strong>the</strong> parameter regions<br />
that <strong>in</strong> <strong>the</strong> skew case correspond to resonance.<br />
2) The regions with code 3 (Hénon-like attractors, red) are quite similar <strong>in</strong> both figures.<br />
3) The region with code 4 <strong>in</strong> Figure 1 (C) (<strong>quasi</strong>-periodic Hénon attractors, light blue),<br />
<strong>in</strong> Figure 6 gives rise to regions <strong>of</strong> codes 4 (light blue) <strong>and</strong> 5 (green) <strong>in</strong> Figure 6, where<br />
<strong>the</strong> difference is given by <strong>the</strong> sign <strong>of</strong> Λ 2 (positive <strong>in</strong> region 4, negative <strong>in</strong> region 5, but<br />
always close to zero).<br />
4) The region with code 2 (blue) <strong>in</strong> Figure 1 (C), where Λ 1 = 0 > Λ 2 has grown smaller<br />
<strong>in</strong> Figure 6. There are blue po<strong>in</strong>ts <strong>in</strong> Figure 1 (C) (not too close to ε = 0) which have<br />
turned <strong>in</strong>to green <strong>in</strong> Figure 6 (Λ 1 > 0 > Λ 2 ). One may expect that <strong>the</strong> dynamics for<br />
30
<strong>the</strong>se parameter values <strong>in</strong> <strong>the</strong> skew case µ = 0 has a <strong>quasi</strong>-periodic attractor. The<br />
numerical evidence, at least work<strong>in</strong>g <strong>in</strong> double precision arithmetics, shows a different<br />
k<strong>in</strong>d <strong>of</strong> attractors <strong>in</strong> <strong>the</strong> fully coupled case µ > 0. In <strong>the</strong> literature <strong>the</strong>se are called<br />
‘strange non-chaotic attractors’ (SNA). See [17, 22, 15, 28, 29, 21] for examples <strong>and</strong><br />
partial results <strong>in</strong> various contexts <strong>and</strong> also Section 4.3. To illustrate this difference,<br />
Figure 10 shows a magnification <strong>of</strong> both Figures 1 (C) <strong>and</strong> 6 for ε ∈ [0, 0.05].<br />
0.04<br />
0.02<br />
0<br />
0.2 0.25 0.3 0.35 0.4<br />
0.04<br />
0.02<br />
0<br />
0.2 0.25 0.3 0.35 0.4<br />
Figure 10: Magnified doma<strong>in</strong> <strong>of</strong> Figures 1 (C) (top) <strong>and</strong> 6 (bottom) show<strong>in</strong>g similarities <strong>and</strong><br />
differences. In <strong>the</strong> top figure we <strong>in</strong>troduced a new code 6 (<strong>in</strong> magenta), located on top <strong>of</strong> <strong>the</strong> blue<br />
region. In Figure 1 (C) this magenta doma<strong>in</strong> was shown <strong>in</strong> blue. It may correspond to ‘strange<br />
non-chaotic attractors’. See <strong>the</strong> text for explanation <strong>and</strong> discussion.<br />
As expla<strong>in</strong>ed <strong>in</strong> Section 4.1 one can use <strong>the</strong> variation as an <strong>in</strong>dicator to dist<strong>in</strong>guish an<br />
<strong>in</strong>variant circle from <strong>in</strong>variant sets <strong>of</strong> o<strong>the</strong>r types. In <strong>the</strong> skew case this reveals a large doma<strong>in</strong><br />
on <strong>the</strong> upper part <strong>of</strong> <strong>the</strong> blue region, <strong>in</strong> Figure 10 represented <strong>in</strong> magenta (code 6). This<br />
is particularly evident near <strong>the</strong> region correspond<strong>in</strong>g to <strong>the</strong> resonance with rotation number<br />
2/7. Narrow doma<strong>in</strong>s can be observed near o<strong>the</strong>r resonances.<br />
Parameter values at <strong>the</strong> top part <strong>of</strong> Figure 10, correspond<strong>in</strong>g to <strong>quasi</strong>-periodic attractors<br />
rema<strong>in</strong> blue (code 2). It is quite strik<strong>in</strong>g that <strong>the</strong>y almost exactly co<strong>in</strong>cide with <strong>the</strong> blue<br />
po<strong>in</strong>ts <strong>in</strong> <strong>the</strong> bottom part <strong>of</strong> <strong>the</strong> figure. This suggests that <strong>the</strong> doma<strong>in</strong> <strong>of</strong> validity <strong>of</strong> <strong>the</strong> kam<br />
Theorem 3 is relatively large.<br />
Even more strik<strong>in</strong>g is that essentially all parameter values with code 6 (say, c<strong>and</strong>idates<br />
for SNA) <strong>in</strong> <strong>the</strong> skew case, enter <strong>in</strong>to <strong>the</strong> ‘green region’ (code 5), with exactly one positive<br />
Lyapunov exponent (where <strong>the</strong> o<strong>the</strong>r two are negative) <strong>in</strong> <strong>the</strong> fully coupled case. This<br />
behaviour has been checked when vary<strong>in</strong>g µ for a sample <strong>of</strong> values <strong>of</strong> (α, ε) : even when µ is<br />
as small as 10 −12 this same change has been observed.<br />
31
4.3 Interpretations <strong>of</strong> <strong>the</strong> numerical results<br />
Most <strong>of</strong> <strong>the</strong> numerical study is based on <strong>the</strong> computation <strong>of</strong> Lyapunov exponents. Know<strong>in</strong>g<br />
<strong>the</strong> values <strong>of</strong> <strong>the</strong>se exponents gives some h<strong>in</strong>ts on <strong>the</strong> dynamics, though certa<strong>in</strong> ambiguities<br />
can occur. As discussed before, <strong>in</strong> cases <strong>of</strong> one positive Lyapunov exponent <strong>and</strong> two negative<br />
ones, one cannot guess whe<strong>the</strong>r <strong>the</strong> attractor is Hénon-like or <strong>quasi</strong>-periodic Hénon-like.<br />
Next we present a couple <strong>of</strong> elementary examples which illustrate how careful one must<br />
be <strong>in</strong> <strong>the</strong> <strong>in</strong>terpretation <strong>of</strong> <strong>the</strong> numerical results.<br />
Arithmetic effects<br />
The computed orbit can strongly depend on <strong>the</strong> k<strong>in</strong>d <strong>of</strong> arithmetics used <strong>in</strong> <strong>the</strong> computations.<br />
Let us return for a moment to <strong>the</strong> Lyapunov sums (35). Assume that <strong>the</strong>y are decreas<strong>in</strong>g<br />
on average <strong>and</strong>, <strong>the</strong>refore, give evidence <strong>of</strong> a negative Lyapunov exponent. However <strong>the</strong><br />
oscillations around a l<strong>in</strong>e with slope equal to <strong>the</strong> Lyapunov exponent can be very large. This<br />
means that local errors can be amplified by a big factor. If <strong>the</strong> amplification is large (before<br />
decreas<strong>in</strong>g aga<strong>in</strong>) <strong>the</strong> propagation <strong>of</strong> <strong>the</strong> numerical errors can show a completely different<br />
dynamics.<br />
To illustrate <strong>the</strong>se numerical effects, let us consider <strong>the</strong> follow<strong>in</strong>g toy model<br />
(x, θ) ↦→ (1 − (a + ε s<strong>in</strong>(2πθ))x 2 , θ + α), (36)<br />
which is <strong>the</strong> Logistic family, driven by a rigid rotation. Note that this is a particular case <strong>of</strong><br />
<strong>the</strong> family (2) for b = δ = 0, which, however, is not a diffeomorphism.<br />
First <strong>of</strong> all we take α small, such that a priori one may expect <strong>the</strong> system (36) to be<br />
not too far from <strong>the</strong> sequence <strong>of</strong> attractors correspond<strong>in</strong>g to <strong>the</strong> ‘frozen’ values <strong>of</strong> θ. We<br />
took (a, ε) = (1, 30, 0.30), <strong>the</strong>reby ensur<strong>in</strong>g that <strong>the</strong> frozen values <strong>of</strong> a + ε s<strong>in</strong>(2πθ) range<br />
over <strong>the</strong> <strong>in</strong>terval [1, 1.6]. In this doma<strong>in</strong> <strong>the</strong> attractors <strong>of</strong> <strong>the</strong> Logistic family range from <strong>the</strong><br />
period 2 s<strong>in</strong>k to <strong>the</strong> chaotic doma<strong>in</strong>, where <strong>the</strong> support <strong>of</strong> <strong>the</strong> <strong>in</strong>variant measure has a s<strong>in</strong>gle<br />
component (say, a ‘one-piece’ strange attractor). The value <strong>of</strong> α has been taken small (<strong>in</strong><br />
particular α = γ/1000, where γ is <strong>the</strong> golden mean) <strong>in</strong> order to move slowly through <strong>the</strong><br />
frozen systems, <strong>in</strong> an adiabatic way.<br />
As a start<strong>in</strong>g po<strong>in</strong>t we chose θ 0 = 0.6. The value <strong>of</strong> x 0 may be taken arbitrary, presently<br />
∑<br />
we picked x 0 = 0.123456789. While comput<strong>in</strong>g iterates <strong>and</strong> <strong>the</strong> Lyapunov sums LS n =<br />
n<br />
j=1 log(2a|x j|) we observe that LS n decreases (with some m<strong>in</strong>or oscillations) dur<strong>in</strong>g <strong>the</strong><br />
first 600 iterates. It reaches a value close to −665. Then LS n <strong>in</strong>creases (except for some m<strong>in</strong>or<br />
oscillations) for <strong>the</strong> next 800 iterates, reach<strong>in</strong>g a value close to −460. This implies that <strong>in</strong> that<br />
transient period, along <strong>the</strong> computed orbit, local errors <strong>in</strong>crease by exp(−460 + 665) ≈ 10 89 .<br />
It may be clear that errors <strong>of</strong> <strong>the</strong> order <strong>of</strong> <strong>the</strong> computer double precision accuracy will<br />
soon produce a departure <strong>of</strong> <strong>the</strong> vic<strong>in</strong>ity <strong>of</strong> <strong>the</strong> orbit that one would obta<strong>in</strong> with exact<br />
computations.<br />
In Figure 11 we present an example <strong>of</strong> this phenomenon. If <strong>the</strong> accuracy is not sufficient,<br />
say only 60 decimal digits, we may expect <strong>the</strong> same to happen, compare with <strong>the</strong> top right<br />
part <strong>of</strong> <strong>the</strong> figure. The lower left plot, computed with 150 decimal digits, displays that<br />
<strong>the</strong> attractor is <strong>in</strong>deed an <strong>in</strong>variant circle. When <strong>the</strong> computer accuracy changes, so does<br />
<strong>the</strong> computed orbit <strong>and</strong> hence also <strong>the</strong> Lyapunov sums change. With <strong>the</strong> ‘correct’ values<br />
<strong>the</strong> first m<strong>in</strong>imum <strong>of</strong> LS n (neglect<strong>in</strong>g period 2 small oscillations) roughly is -646 atta<strong>in</strong>ed<br />
at n = 562. After that we obta<strong>in</strong> a maximum <strong>of</strong> -376 for n = 1459. The magnification<br />
exp(−376 + 646) ≈ 10 117 shows why a large number <strong>of</strong> digits is required.<br />
For frozen θ <strong>the</strong> effective value <strong>of</strong> <strong>the</strong> parameter <strong>in</strong> <strong>the</strong> Logistic family is â = a+ε s<strong>in</strong>(2πθ),<br />
see (36). Period two po<strong>in</strong>ts <strong>of</strong> <strong>the</strong> frozen system are born at â = 3/4, as solutions <strong>of</strong> â 2 x 2 −<br />
32
1<br />
1<br />
0.5<br />
0.5<br />
0<br />
0<br />
-0.5<br />
-0.5<br />
0 0.2 0.4 0.6 0.8 1<br />
0 0.2 0.4 0.6 0.8 1<br />
1<br />
0<br />
0.5<br />
-1000<br />
-2000<br />
0<br />
-3000<br />
-0.5<br />
0 0.2 0.4 0.6 0.8 1<br />
-4000<br />
0 2000 4000 6000 8000 10000<br />
Figure 11: Top <strong>and</strong> bottom left: attractors observed for <strong>the</strong> model (36) with (a, ε, α) =<br />
(1.30, 0.30, γ/1000), where γ denotes <strong>the</strong> golden mean. In <strong>the</strong> top part we use st<strong>and</strong>ard double<br />
precision arithmetics (left) <strong>and</strong> 60 decimal digit arithmetics (right). The bottom left figure uses<br />
150 decimal digit arithmetics. The bottom right part depicts <strong>the</strong> correspond<strong>in</strong>g evolution <strong>of</strong> <strong>the</strong><br />
Lyapunov sums (35). In <strong>the</strong> present example <strong>in</strong>creas<strong>in</strong>g <strong>the</strong> accuracy also gives an <strong>in</strong>crease <strong>of</strong> <strong>the</strong><br />
Lyapunov exponent, but it rema<strong>in</strong>s negative <strong>in</strong> all three <strong>the</strong> cases. See <strong>the</strong> text for details.<br />
âx+1−â = 0. On <strong>the</strong> correspond<strong>in</strong>g orbit one has |D 2 T (x)| = 4|(â−1)|. Hence, <strong>the</strong> Lyapunov<br />
sums (35) are Riemann sums <strong>of</strong> <strong>the</strong> <strong>in</strong>tegral<br />
∫<br />
1<br />
θ0 +nα<br />
log (|4(a − 1 + ε s<strong>in</strong>(2πθ))|) dθ. (37)<br />
2α θ 0<br />
We note that <strong>the</strong> <strong>in</strong>tegral (37) cannot be dist<strong>in</strong>guished from <strong>the</strong> upper curve at <strong>the</strong> bottom<br />
right part <strong>of</strong> Figure 11.<br />
Invariant curves with large oscillations<br />
As mentioned before, one might expect that an <strong>in</strong>variant curve looks very nice, with moderate<br />
oscillation, <strong>and</strong> decide that a wild oscillation should be a good evidence <strong>of</strong> <strong>the</strong> existence <strong>of</strong><br />
an SNA. The follow<strong>in</strong>g example shows that this expectation is not justified <strong>in</strong> all cases.<br />
We consider a forced Logistic family, as studied <strong>in</strong> [27], given by<br />
(x, θ) ↦→ (µx(1 − x) + ε s<strong>in</strong>(2πθ)), θ + α(mod 1)) , (38)<br />
where µ = 3 <strong>and</strong> α = γ (<strong>the</strong> golden mean). In [27] <strong>the</strong> authors claim that start<strong>in</strong>g at<br />
ε = ε ∗ , where ε ∗ ≈ 0.1553, <strong>the</strong>re exists a range <strong>of</strong> values <strong>of</strong> ε display<strong>in</strong>g an SNA. Despite<br />
<strong>the</strong> Lyapunov exponent <strong>in</strong> <strong>the</strong> x variable is negative <strong>the</strong> attractors look like ‘strange’, hav<strong>in</strong>g<br />
fractal dimension. Their conclusions are based on a too small number <strong>of</strong> iterates.<br />
To clarify <strong>the</strong> dynamics we have proceeded as follows: We computed, after some transient,<br />
N iterates <strong>of</strong> (38). These values were sorted with respect to θ <strong>and</strong> <strong>the</strong> oscillations were<br />
computed <strong>of</strong> <strong>the</strong> x variable <strong>in</strong> <strong>the</strong> <strong>in</strong>tervals [0.0, 0.1], . . . , [0.9, 1.0]. Let Ij 1 1<br />
:= [j 1 /10, (j 1 +1)/10]<br />
be <strong>the</strong> <strong>in</strong>terval with largest oscillation. Then we recompute 10N iterates consider<strong>in</strong>g only <strong>the</strong><br />
33
iterates <strong>in</strong> Ij 1 1<br />
. These iterates are sorted with respect to θ <strong>and</strong> <strong>the</strong> oscillations <strong>of</strong> <strong>the</strong> x variable<br />
are computed <strong>in</strong> <strong>the</strong> sub<strong>in</strong>tervals <strong>of</strong> <strong>the</strong> form [j 1 /10+k/100, j 1 /10+(k+1)/100], k = 0, 1, . . . 9.<br />
Let Ij 2 1 ,j 2<br />
be <strong>the</strong> sub<strong>in</strong>terval with largest oscillation. This process is repeated as many times<br />
as needed, until <strong>the</strong> maximal slope, based on <strong>the</strong> computed po<strong>in</strong>ts, is no longer chang<strong>in</strong>g <strong>in</strong> a<br />
significant way. Figure 12(left) shows <strong>the</strong> results for ε = 0.1554. The observed attractor, that<br />
with smaller resolution may seem a strange attractor, is <strong>in</strong> fact a nice curve, certa<strong>in</strong>ly with a<br />
large oscillation <strong>and</strong> with large slope. Due to <strong>the</strong> fact that <strong>the</strong> method computes oscillations<br />
based on a grid, <strong>the</strong> f<strong>in</strong>ally selected <strong>in</strong>terval may depend on <strong>the</strong> value <strong>of</strong> N <strong>and</strong> on <strong>the</strong> <strong>in</strong>itial<br />
values x 0 , θ 0 , but <strong>the</strong> results are similar. Of course, due to <strong>the</strong> large number <strong>of</strong> iterations<br />
performed, <strong>the</strong>re is a loss <strong>of</strong> digits. To overcome this source <strong>of</strong> errors we have used between<br />
30 <strong>and</strong> 40 decimal digits <strong>in</strong> <strong>the</strong> computations.<br />
0.8<br />
0.8<br />
0.7<br />
0.7<br />
0.6<br />
0.6<br />
0.5<br />
0 2000 4000 6000 8000 10000<br />
0.5<br />
0 10000 20000 30000 40000 50000<br />
Figure 12: Invariant curves for map (38). Left: ε = 0.1554. On <strong>the</strong> horizontal axis we plot<br />
(θ − 0.0070944247) × 10 14 <strong>and</strong> on <strong>the</strong> vertical axis we plot x. Right: ε = 0.1555. On <strong>the</strong> horizontal<br />
axis we plot (θ − 0.007235958375) × 10 16 . The maximum slopes <strong>in</strong> <strong>the</strong> left <strong>and</strong> right plots are<br />
1.5 × 10 12 <strong>and</strong> 4.0 × 10 14 , respectively.<br />
In <strong>the</strong> righth<strong>and</strong> part <strong>of</strong> Figure 12 similar results are shown for ε = 0.1555, obta<strong>in</strong>ed by<br />
us<strong>in</strong>g a variety <strong>of</strong> different methods. Prelim<strong>in</strong>ary results give evidence that for ε = 0.1556<br />
<strong>the</strong> largest slope exceeds 10 18 . Hav<strong>in</strong>g negative Lyapunov exponent <strong>in</strong> <strong>the</strong> x variable implies<br />
that <strong>the</strong> cont<strong>in</strong>uation <strong>of</strong> <strong>the</strong> <strong>in</strong>variant curve with respect to ε is still locally possible. For<br />
fur<strong>the</strong>r examples <strong>and</strong> <strong>the</strong>oretical discussion we refer to [21].<br />
Summaris<strong>in</strong>g, we conclude that certa<strong>in</strong> phenomena which might be attributed to <strong>the</strong><br />
dynamics can, <strong>in</strong> fact, be due to a wrong <strong>in</strong>terpretation <strong>of</strong> <strong>the</strong> results or to computations<br />
done with too few digits. This does not mean that results obta<strong>in</strong>ed with a fewer number<br />
<strong>of</strong> digits are not important. Indeed, most <strong>of</strong> <strong>the</strong> ma<strong>the</strong>matical models used for concrete<br />
applications are approximations <strong>and</strong>, fur<strong>the</strong>rmore, ‘real life’ problems always conta<strong>in</strong> some<br />
amount <strong>of</strong> noise. The role played <strong>in</strong> <strong>the</strong>se toy models by <strong>the</strong> round<strong>in</strong>g errors can be viewed<br />
as noise. So <strong>the</strong> behaviour <strong>of</strong> a real system can be closer to <strong>the</strong> top left <strong>of</strong> Figure 11 ra<strong>the</strong>r<br />
than to <strong>the</strong> bottom left one. But it is always better to known why.<br />
References<br />
[1] V.I. Arnol ′ d: Small denom<strong>in</strong>ators, I: Mapp<strong>in</strong>gs <strong>of</strong> <strong>the</strong> circumference <strong>in</strong>to itself, AMS<br />
Transl. (Ser. 2) 46 (1965), 213–284.<br />
[2] M. Benedicks, L. Carleson: On iterations <strong>of</strong> 1−ax 2 on (−1, 1), Ann. <strong>of</strong> Math. (2) 122(1)<br />
(1985), 1–25.<br />
34
[3] M. Benedicks, L. Carleson: The dynamics <strong>of</strong> <strong>the</strong> Hénon map, Ann. <strong>of</strong> Math. (2) 133(1)<br />
(1991), 73–169.<br />
[4] M. Bosch, J.P. Carcassès, C. Mira, C. Simó, J.C. Tatjer: “Crossroad area-spr<strong>in</strong>g area”<br />
transition. (I) Parameter plane representation. Int. J. <strong>of</strong> Bifurcation <strong>and</strong> <strong>Chaos</strong> 1(1)<br />
(1991), 183–196.<br />
[5] H.W. Broer, G.B. Huitema, M.B. Sevryuk: Quasi-periodic Motions <strong>in</strong> Families <strong>of</strong> Dynamical<br />
Systems, Order amidst <strong>Chaos</strong>, Spr<strong>in</strong>ger LNM 1645 (1996).<br />
[6] H.W. Broer, G.B. Huitema, F. Takens, B.L.J. Braaksma: Unfold<strong>in</strong>gs <strong>and</strong> bifurcations <strong>of</strong><br />
<strong>quasi</strong>-periodic tori, Mem. AMS 83(421) (1990), 1–175.<br />
[7] H.W. Broer, C. Simó: Hill’s equation with <strong>quasi</strong>–periodic forc<strong>in</strong>g: resonance tongues,<br />
<strong>in</strong>stability pockets <strong>and</strong> global phenomena. Bul. Soc. Bras. Mat. 29 (1998), 253–293.<br />
[8] H.W. Broer, C. Simó, J.C. Tatjer: Towards global models near homocl<strong>in</strong>ic tangencies <strong>of</strong><br />
dissipative <strong>diffeomorphisms</strong>, Nonl<strong>in</strong>earity 11 (1998), 667–770.<br />
[9] H.W. Broer, C. Simó, R. Vitolo: Bifurcations <strong>and</strong> strange attractors <strong>in</strong> <strong>the</strong> Lorenz-84<br />
climate model with seasonal forc<strong>in</strong>g, Nonl<strong>in</strong>earity 15(4) (2002), 1205–1267.<br />
[10] H.W. Broer, C. Simó, R. Vitolo: Quasi-periodic Hénon-like attractors <strong>in</strong> <strong>the</strong> Lorenz-84<br />
climate model with seasonal forc<strong>in</strong>g, to appear <strong>in</strong> Proceed<strong>in</strong>gs Equadiff 2003.<br />
[11] H.W. Broer, F. Takens: Formally symmetric normal forms <strong>and</strong> genericity, Dynamics<br />
Reported 2 (1989), 36–60.<br />
[12] P.M. C<strong>in</strong>cotta, C.M. Giordano, C. Simó: Phase space structure <strong>of</strong> multidimensional<br />
systems by means <strong>of</strong> <strong>the</strong> Mean Exponential Growth factor <strong>of</strong> Nearby Orbits (MEGNO),<br />
Physica D 182 (2003), 151–178.<br />
[13] R.L. Devaney: An Introduction to Chaotic Dynamical Systems (2nd edition), Addison–<br />
Wesley (1989).<br />
[14] L. Díaz, J. Rocha, M. Viana: Strange attractors <strong>in</strong> saddle cycles: prevalence <strong>and</strong> globality,<br />
Inv. Math. 125 (1996), 37–74.<br />
[15] P. Glend<strong>in</strong>n<strong>in</strong>g: Intermittency <strong>and</strong> strange nonchaotic attractors <strong>in</strong> <strong>quasi</strong>-periodically<br />
forced circle maps, Phys. Lett. A 244 (1998), 545–550.<br />
[16] S.V. Gonchenko, I.I. Ovsyannikov, C. Simó, D. Turaev: Three-dimensional Hénon maps<br />
<strong>and</strong> wild Lorenz–type strange attractors, prepr<strong>in</strong>t, 2005.<br />
[17] C. Grebogi, E. Ott, S. Pelikan, J. Yorke: Strange attractors that are not chaotic, Physica<br />
D 13(1-2) (1984), 261–268.<br />
[18] M. Hénon: A two dimensional mapp<strong>in</strong>g with a strange attractor, Comm. Math. Phys.<br />
50 (1976), 69–77.<br />
[19] M.W. Hirsch: Differential Topology, Spr<strong>in</strong>ger GTM 33 (1976).<br />
[20] M.W. Hirsch, C.C. Pugh, M. Shub: Invariant Manifolds, Spr<strong>in</strong>ger LNM 583 (1977).<br />
[21] À. Jorba, J.C. Tatjer: On <strong>the</strong> fractalization <strong>of</strong> <strong>in</strong>variant curves <strong>in</strong> <strong>quasi</strong>-periodically<br />
forced 1-D systems, prepr<strong>in</strong>t, 2005.<br />
35
[22] G. Keller: A note on strange nonchaotic attractors, Fund. Math. 151(2) (1996), 139–148.<br />
[23] F. Ledrappier, M. Shub, C. Simó, A. Wilk<strong>in</strong>son: R<strong>and</strong>om versus determ<strong>in</strong>istic exponents<br />
<strong>in</strong> a rich family <strong>of</strong> <strong>diffeomorphisms</strong>, J. <strong>of</strong> Stat. Phys. 113 (2003), 85–149.<br />
[24] W. de Melo, S. van Strien: One dimensional dynamics, Spr<strong>in</strong>ger–Verlag (1993).<br />
[25] M. Misiurewicz: Absolutely cont<strong>in</strong>uous measures for certa<strong>in</strong> maps <strong>of</strong> an <strong>in</strong>terval, Publ.<br />
Math. IHES 53 (1981), 17–51.<br />
[26] L. Mora, M. Viana: Abundance <strong>of</strong> strange attractors, Acta Math 171 (1993), 1–71.<br />
[27] T. Nishikawa, K. Kaneko: Fractalization <strong>of</strong> a <strong>torus</strong> as a strange nochaotic attractor,<br />
Phys. Rev. E 56(6) (1996), 6114–6124.<br />
[28] H. Os<strong>in</strong>ga, U. Feudel: Boundary crisis <strong>in</strong> <strong>quasi</strong>periodically forced systems, Physica D<br />
141(1-2) (2000), 54–64.<br />
[29] H. Os<strong>in</strong>ga, J. Wiersig, P. Glend<strong>in</strong>n<strong>in</strong>g, U. Feudel: Multistability <strong>in</strong> <strong>the</strong> <strong>quasi</strong>periodically<br />
forced circle map, IJBC 11 (2001), 3085–3105.<br />
[30] J. Palis, W. de Melo: Geometric Theory <strong>of</strong> Dynamical Systems, An Introduction<br />
Spr<strong>in</strong>ger–Verlag (1982).<br />
[31] J. Palis, F. Takens: Hyperbolicity & Sensitive Chaotic Dynamics at Homocl<strong>in</strong>ic Bifurcations,<br />
Cambridge Studies <strong>in</strong> Advanced Ma<strong>the</strong>matics 35, Cambridge University Press<br />
(1993).<br />
[32] J. Palis, M. Viana: High dimension <strong>diffeomorphisms</strong> display<strong>in</strong>g <strong>in</strong>f<strong>in</strong>itely many periodic<br />
attractors, Ann. <strong>of</strong> Math. (2) 140(1) (1994), 91–136.<br />
[33] M. Shub: Global stability <strong>of</strong> dynamical systems, Spr<strong>in</strong>ger–Verlag (1986).<br />
[34] C. Simó: On <strong>the</strong> Hénon-Pomeau attractor, J. <strong>of</strong> Stat. Phys. 21 (1979), 465–494.<br />
[35] C. Simó: On <strong>the</strong> use <strong>of</strong> Lyapunov exponents to detect global properties <strong>of</strong> <strong>the</strong> dynamics,<br />
to appear <strong>in</strong> Proceed<strong>in</strong>gs Equadiff03.<br />
[36] J.C. Tatjer: Three dimensional dissipative <strong>diffeomorphisms</strong> with homocl<strong>in</strong>ic tangencies,<br />
ETDS 21(1) (2001), 249–302.<br />
[37] Ph. Thieullen, C. Tresser, L.-S. Young: Positive Lyapunov exponent for generic oneparameter<br />
families <strong>of</strong> unimodal maps, J. Anal. Math. 64 (1994), 121–172.<br />
[38] M. Tsujii: A simple pro<strong>of</strong> for monotonicity <strong>of</strong> entropy <strong>in</strong> <strong>the</strong> quadratic family, ETDS<br />
20 (2000), 925–933.<br />
[39] M. Viana: Strange attractors <strong>in</strong> higher dimensions, Bol. Soc. Bras. Mat 24 (1993),<br />
13–62.<br />
[40] M. Viana: Multidimensional nonhyperbolic attractors, Publ. Math. IHES 85 (1997),<br />
63–96.<br />
[41] R. Vitolo: Bifurcations <strong>of</strong> attractors <strong>in</strong> 3D <strong>diffeomorphisms</strong>, PhD <strong>the</strong>sis, University <strong>of</strong><br />
Gron<strong>in</strong>gen (2003).<br />
[42] Q. Wang, L.-S. Young: Strange Attractors with One Direction <strong>of</strong> Instability, Comm.<br />
Math. Phys. 218 (2001), 1–97.<br />
36